Uncovering an Adipocyte's Perspective of Inflammation And

Uncovering an Adipocyte’s Perspective of

Inflammation and Immunity in Obesity

A dissertation submitted to the Graduate School of the University of Cincinnati

In partial fulfillment of the requirements for the degree of

DOCTOR OF PHILOSOPHY

Immunology Graduate Program, College of Medicine

2019

Calvin C. Chan

B.S., University of Mary Washington, 2010

M.S., University of Cincinnati, 2013

Committee Chair: Senad Divanovic, Ph.D.

ABSTRACT

Adipocytes, traditionally perceived to be simple energy storing cells, are highly complex biosynthetic factories that regulate overall metabolic health. Among their multi-potent capacities, evidence hints at an intertwined relationship between adipocytes and inflammation. Like immune cells, adipocytes are highly plastic and consists of different subsets, including white, brown, beige and pink adipocytes. Adipocytes can conduct themselves in an “immune-like” manner and inflammation can influence adipocyte lipid metabolism. However, mechanisms underpinning adipocyte inflammatory potential and inflammatory regulation of adipocyte homeostasis remain critical gaps in knowledge. As long underappreciated cells, elucidating this complex connection holds promise to uncover how adipocyte-inflammation integrates within human health and disease.

In this dissertation, we have identified two inflammatory pathways (type I interferon

(IFN)/interferon αβ receptor (IFNAR) and BAFF/APRIL axes) that regulate adipocyte functional potential. Our data define the role of type I IFN/IFNAR axis in unleashing dormant adipocyte inflammatory capacity. Notably, activation of IFNAR promotes an

“immune-like” transcriptome and amplifies a glycolysis-associated inflammatory vigor in adipocytes. We posited that type I IFN/IFNAR-driven adipocyte inflammatory vigor could impact disease pathogenesis, including obesity. Indeed, obesity not only augments the type I IFN axis in adipocytes but also enhanced their responsiveness to type I IFN effects.

Notably, non-hematopoietic (e.g., adipocytes) IFNAR expression contributed to obesity- associated metabolic dysfunction. These findings highlight for the first time a mechanism revealing adipocyte pathogenic inflammatory potential and its potential contribution in the context of obesity.

i As adipocytes and inflammation are recognized pathogenic agents in the context of obesity, we initially sought to define whether deletion of a negative regulator of TLR activity, radioprotective 105 (RP105), would exacerbate adipocyte and inflammation pathogenic effects. This pursuit led to the serendipitous discovery of the B cell activating factor (BAFF)/A proliferation inducing ligand (APRIL) axis-mediated protection from obesity development. Multiple transgenic mouse lines exhibiting increased systemic

BAFF levels were protected from diet-induced obesity (DIO). While genetic deletion of

BAFF or APRIL alone was insufficient to reverse resistance to DIO, removal of both BAFF and APRIL fully exacerbated diet-driven weight gain. Dissection of the source of this protection revealed that both BAFF and APRIL modified white adipocyte lipolysis, the breakdown of fat, increased brown adipocyte thermogenesis, and total body energy expenditure. Higher systemic BAFF or APRIL levels were directly correlated with greater weight loss after bariatric surgery. Thus, these findings highlight BAFF/APRIL as beneficial and multifaceted inflammatory mediators that may play a vital role in obesity development.

Overall, our findings shed light on an intricate balance between inflammation and adipocyte function. Harnessing our understanding of how to modulate the detrimental

(inflammatory vigor) and beneficial (lipid/energy handling) aspects of the adipocyte- inflammation axis may yield insights into previously hidden therapeutic avenues for human health and disease.

iii

ACKNOWLEDGEMENTS

This great accomplishment could not have happened without wonderful and wholeheartedly supportive individuals in my life.

I sincerely thank my mentor Senad whose guiding hand has shaped me into an exponentially better scientist. Your daily enthusiasm and scientific curiosity are infectious and I thank you for nurturing my scientific passion and for feeding my scientific curiosities.

Your mentorship has helped me develop my most valuable tools, the ability to interpret science through a critical lens and to continue asking the most impactful questions. Thank you for believing in my potential and thank you for molding that potential into a better, more complete person.

To all of the past and current members of the Divanovic laboratory, I thank you for your unyielding support and constant joy inside and outside of the lab. Special thanks to

Traci and Maru for your guidance early on which helped set me on the path to becoming a better scientist and person. Many thanks to Dan, Monica, Michelle, Jarren, Matt, Rajib,

Pablo, and Jessica for all of your time, input, and help with this body of work. It was an absolute pleasure working with each and every one of you.

I would also like to extend my gratitude to my dissertation committee. George,

Kasper, Ian, and Tom. You have all provided your utmost support from day one and have constantly pushed me to think outside of the box. Your efforts have helped me grow immensely and I have learned to fully appreciate “the big picture” and how to pursue high impact science.

My deepest appreciation for my parents May and Danny. You have been the best, most loving parents that have taught me the value of being curious, being persistent, and

iv working hard every day. I am a better son, husband, and father because of you. Both of you have been the biggest inspiration in my life and I dedicate this dissertation to you – the fruitful outcome of all of your efforts. I am also very grateful to my sister Carmen. You have shown me that true success in life feels more like a marathon than a sprint. The time and dedication you have put into become a physician has been a true inspiration. Your support over these many years has been appreciated each and every day.

Finally, my deepest gratitude goes to my wife Maegan. Your unconditional love, support, encouragement has been irreplaceable. You have created a safe space for me outside of science and empowered me to constantly pursue lofty goals. You have pushed me forward towards achieving my dreams when in times I felt like they were far far away.

I also dedicate this body of work to my newborn son Liam; you have been a spark of joy every day. I absolutely look forward to our adventures ahead.

ABBREVIATIONS AND ACRYONMS

AT, Adipose Tissue WAT, White adipose tissue BAT, Brown adipose tissue HFD, High-fat diet CD, Chow diet (low-fat) DIO, Diet-induced obesity APC, Antigen presenting cell DC, Dendritic cell MHC, Major Histocompatability Complex Th, T helper cell CCL, C-C motif chemokine ligand CXCCL, C-X-C motif chemokine ligand LIPE, Hormone Sensitive Lipase PNPLA2, Adipose Triglyceride Lipase FABP4, Fatty acid binding protein 4 ADIPOQ, Adiponectin IFN, Interferon IFNAR, Interferon α/β receptor IL, Interleukin TNF, Tumor Necrosis factor TLR, Toll-like receptor NLR, Nod-like receptor MAVS, Mitochondrial antiviral-signaling protein STING, Stimulator of interferon genes RLR, RIG-I Receptor BAFF, B cell activating factor APRIL, A proliferation-inducing ligand BAFF-R, B cell activating factor-receptor TACI, Transmembrane activator and CAML interactor

BCMA, B-cell maturation antigen HK, Hexokinase PGK, Phosphoglycerate kinase PK, Pyruvate kinase PCR, Polymerase chain reaction WB, Western Blot FACS, Fluorescence-activated cell sorter RNA-seq, RNA sequencing ATAC-seq, Assay for Transposase-Accessible Chromatin sequencing LCMS, Liquid chromatography mass spectroscopy NMR, Nuclear magnetic resonance Met-H, Metabolically healthy Met-C, Metabolically challenged LAL, Limulus amebocyte lysate assay GTT, Glucose tolerance test ITT, Insulin tolerance test ALT, Alanine transaminase AST, Aspartate transaminase GGT, Gamma-Glutamyl Transferase NAS, NAFLD activity score HOMA-IR, Homeostatic model assessment-Insulin resistance TG, Triglycerides FFA, Free fatty acid OCR, Oxygen consumption rate ECAR, Extracellular acidification rate SLE, Systemic Lupus Erythematosus 2-dg, 2-deoxyglucose BMI, Body mass index

vii

TABLE OF CONTENTS

ABSTRACT ...... i

ACKNOWLEDGEMENTS ...... iv

ABBREVIATIONS AND ACRYONMS ...... vi

TABLE OF CONTENTS ...... viii

Chapter 1. Inflammation and immunity: From an adipocyte’s perspective ...... 1 ADIPOCYTES ...... 4 ADIPOCYTE-INTRINSIC “IMMUNE-LIKE” POTENTIAL ...... 6 INFLAMMATION AND ALTERATION OF ADIPOCYTE FUNCTION ...... 11 ADIPOCYTE-INFLAMMATION AXIS IN DISEASE PATHOGENESIS ...... 15

Chapter 2. Obesity and its Association with Inflammation ...... 33 OBESITY PANDEMIC ...... 33 OVERVIEW OF TYPE I INTERFERONS ...... 36 OVERVIEW OF B CELL ACTIVATING FACTOR (BAFF) AND A PROLIFERATION-INDUCING LIGAND (APRIL) AXIS ...... 43

Chapter 3. Type I Interferon Sensing Unlocks Dormant Adipocyte Inflammatory Potential ...... 57

Chapter 4. A BAFF/APRIL axis regulates obesogenic-diet driven weight gain ...... 117

Chapter 5. Conclusions, Discussion and Future Directions ...... 174 TYPE I IFN/IFNAR CONTROL OF ADIPOCYTE INFLAMMATORY VIGOR ...... 175 BAFF/APRIL AXIS CONTROL OF ADIPOCYTE LIPID HANDLING ...... 188 TYPE I IFN/IFNAR AND BAFF-APRIL BALANCE IN ADIPOCYTE INFLAMMATION ...... 196 SUMMARY: OVERARCHING POTENTIAL EXPLOITATION OF ADIPOCYTE-INFLAMMATION AXIS IN HUMAN HEALTH AND DISEASE ...... 198

APPENDIX ...... 204

viii

Chapter 1. Inflammation and immunity: From an adipocyte’s perspective

Calvin C. Chan1,2,3, Michelle S.M.A. Damen2,3, Pablo C. Alarcon1,2,3, Joan Sanchez- Gurmaches2,4,5, Senad Divanovic1,2,3

*Correspondence: Senad Divanovic, Department of Pediatrics, University of Cincinnati College of Medicine, Division of Immunobiology, Cincinnati Children's Hospital Medical Center, Cincinnati, OH. TCHRF - Location S, Room #S.5.409 3333 Burnet Avenue, Cincinnati, Ohio 45229-3039 U.S.A. Phone: 513-636-0286, Fax: 513-636-5355, e-mail: [email protected].

Running Title: Adipocytes and Inflammation

Keywords: Adipocytes, Inflammation, Cytokines, Interferon, Metabolism, Obesity, BAFF

1 ABSTRACT

Comprehension of adipocyte function has evolved beyond a long-held belief of their inert nature, as simple energy storing and releasing cells. Adipocytes, including white, brown and beige, are capable mediators of global metabolic health, but their intersection with inflammation is a budding field of exploration. Evidence hints at a reciprocal relationship adipocyte share with immune cells. Adipocyte’s capacity to behave in an “immune-like” manner and ability to sense inflammatory cues that subsequently alter core adipocyte function might play an important role in shaping immune responses. Clarifying this intricate relationship could uncover previously underappreciated contribution of adipocytes to inflammation-driven human health and disease. Here, we highlight the potential of largely underappreciated adipocyte “immune-like” function and how it may contribute to inflammation, immunity and pathology of various diseases.

INTRODUCTION

Dogma has long suggested that adipose tissue (AT), mainly described to consist of white adipose tissue (WAT) and brown adipose tissue (BAT), was a simplistic organ. The common understanding is that WAT plays a central role in energy storage, while BAT is tied to energy expenditure via adaptive non-shivering thermogenesis. While research focused on AT biology and function had stayed relatively dormant, the obesity pandemic has reignited research interests into AT biology and its contribution to disease pathogenesis. In fact, the recently uncovered appreciation of intricate biological processes that govern AT development, function and contribution to health and disease, makes the AT enigmatic function a highly attractive research area.

AT houses a rich milieu of cells, including adipocytes, preadipocytes (progenitor stem cells within the stromal vascular fraction), fibroblasts, endothelial cells and immune cells, which collectively play an important role in the metabolic regulation/whole-body energy homeostasis1,2. Sensing of environmental homeostatic (e.g., energy flux) or disease (e.g., obesity, infection, cancer) cues by AT alter its composition to adapt and contribute, either in beneficial or detrimental means1,3-8. In fact, recent discoveries revealing that AT is a highly plastic organ with dynamic immune-like characteristics6,9-11 further support the call for a better functional understanding of AT constituents in homeostasis and disease pathogenesis.

The dogma surrounding AT inflammatory capacity suggests that immune cells infiltrating AT are the primary source of inflammation. Multiple published reviews have previously focused on the contribution of WAT infiltrating immune cells (e.g., macrophages, dendritic cells [DC], T cells, B cells, natural killer [NK] cells)12,13 and hence

3 will not be discussed here. Although underdefined and an active area of investigation, compared to WAT, the immune cell milieu is also present in reduced numbers in BAT and beige AT (e.g., macrophages, T, and B cells)11,14,15. Notably, traditional approaches to study AT inflammation have almost entirely focused on examination of WAT as a whole or specific WAT infiltrating immune cells. Importantly however, the intrinsic contribution of adipocytes and/or their potential role in the establishment and modulation of the overall

AT inflammatory capacity is underappreciated and has not been well discussed. Hence, this review will summarize the recently established landscape of adipocyte contribution to AT inflammatory capacity (e.g., cytokine/chemokine production, immune cell crosstalk) and will invoke critical future directions that should be investigated to broaden our understanding of this topic.

ADIPOCYTES

A fundamental feature of adipocytes, which form the core of AT, is nutrient handling, energy storage and utilization, and secretion of adipokines (e.g., leptin, adiponectin, resistin) that can modify total body metabolic homeostasis10. Adipocytes are thus appreciated to be highly complex biosynthetic factories. Currently, adipocytes can be divided in multiple subtypes: (a) white adipocytes (energy storing); (b) brown adipocytes

(energy expending); (c) beige adipocytes (also called brite adipocyte; inducible, energy expending); and (d) pink adipocytes (lipid-rich, pregnancy/lactation associated) (Figure

1). Remarkable adipocyte plasticity, which remains an exciting area of investigation, has been extensively discussed elsewhere9,16, hence will not be reviewed here.

White adipocytes regulate the assimilation of excess energy (e.g., circulating dietary lipids) for storage and in part through de novo lipogenesis, resulting in increased lipid droplet size and expansion of the AT17. In contrast, when shortage of nutrients occurs, adipocytes can shrink in size by hydrolyzing stored triglycerides (TG) into fatty acids (FA) and glycerol through lipolysis18 to provide other organs/cells sufficient energy

19. Brown adipocytes consist of numerous mitochondria and are highly active metabolic cells that have the capacity for adaptive non-shivering thermogenesis, the process of using energy to produce heat necessary for temperature regulation20. Beige adipocytes constitute inducible thermogenic cells that arise within WAT in response to environmental cues (e.g., cold stress, exercise)20,21. Beige adipocytes are derived from distinct progenitors and potentially from transdifferentiation of white adipocytes. However, technical limitations, including lack of a specific marker for differentiated white adipocytes, prohibit precise clarity of transdifferentiation between white and beige adipocytes and further investigation is needed20. The recently discovered pink adipocytes appear during pregnancy and are intricately involved in post-pregnancy lactation. Arising from subcutaneous white adipocytes within the mammary glands, pink adipocytes are a lipid rich energy source for the production of milk22. Given their recent discovery, our understanding of the intersection between inflammation and pink adipocytes is limited22.

While an exciting and novel avenue of investigation, pink adipocytes will not be further discussed within this review.

The ability to generate and manipulate primary adipocytes from preadipocytes1,2,23 has enhanced our understanding of the multifactorial impact adipocytes themselves may play in the establishment and modulation of the overall AT immune capacity. Ex vivo

5 derived adipocyte differentiation, from preadipocytes, is a highly complex process wherein activation of peroxisome proliferator-activated receptor gamma (PPAR-γ),

CCAAT/enhancer-binding proteins (C/EBP) and sterol regulatory element binding protein (SREBP1) drive an adipocyte fate2,23. Despite the ability to generate ex vivo adipocytes from preadipocytes, their direct comparison to AT isolated mature adipocytes remains an area of intense investigation23.

Representing the specific focus of this review, functional potential for adipocytes to behave like immune cells has recently been invoked. These provocative studies suggest that the breadth of the adipocyte “immune-like” capacity ranges from their ability to express innate immune receptors24-28, produce proinflammatory cytokines26,28-31, express chemokines32, and present antigens3,33-37 (Figure 2). Such findings highlight a newfound intricacy and potential for adipocytes to modulate AT and systemic inflammation that modify global immune responsiveness under homeostatic and disease state conditions.

ADIPOCYTE-INTRINSIC “IMMUNE-LIKE” POTENTIAL

Investigations, primarily utilizing mature or preadipocyte-derived primary adipocytes, has uncovered that adipocytes express various innate immune receptors including Toll-like receptors (TLR), NOD-like receptors (NLR), and RIG-I-like receptors (RLR). Screening of

TLRs within mouse or human white adipocytes have revealed their expansive presence within these cells. As with immune cells38, human white adipocytes express all 10 known

TLRs with TLR3, TLR4, TLR5 and TLR9 being the most highly expressed24. In contrast, of 12 known TLRs38, mouse white adipocytes express TLR1, TLR2, TLR3, TLR4, TLR5,

TLR6, TLR7 and TLR925,27,28. Further, expression of TLR2 and TLR4 in mouse brown adipocytes has been similarly detected39. In addition, of the 34 NLRs in mice40, white adipocytes are known to express NOD-1, NOD-2 and NLRP341-43 while only NOD-1 is detected in brown adipocytes39. NOD-1, but not NOD-2, suppress white adipocyte differentiation and induces the NF-kB pathway to activate innate responses42,43. Despite detectable NLRP3 expression in white adipocytes, a critical component in the inflammasome that leads to caspase-1 activation41, NLRP3 expression is low compared to macrophages thereby raising questions of its role and contribution in adipocytes. The finite knowledge surrounding adipocyte NLR expression and contribution obviously warrants future studies formally interrogating this axis. White adipocytes also express

RIG-I44, a sensor of double stranded RNA and viruses, but their functionality remains inconclusive and formal examination on adipocyte “immune-like” function and relevance to viral infections is needed. In addition, given their relevance to shaping immune responses, studies focused on the expression and role of C-type lectin receptors (CLR) and AIM2-like receptors (ALR) in adipocyte (white, brown, beige) “immune-like potential” are warranted.

To date, expression and contribution of TLRs to adipocyte inflammatory potential has been best studied. Adipocyte sensing of well-established, TLR recognized, pathogen associated molecular patterns (PAMPs) (e.g., LPS, zymosan, poly I:C, flagellin, loxoribine, CpG) by both mature and preadipocyte-derived white adipocytes results in production of multiple proinflammatory mediators26,28,31. As obesity is associated with increased systemic endotoxemia45, circulating DNA46 and likely other PAMPs/danger associated molecular patterns (DAMPs), these findings suggest that adipocyte-intrinsic

7 innate immune receptor activation may also contribute to the overall tissue and systemic inflammatory environment in obesity or other diseases associated with similar disturbances (e.g., inflammatory bowel disease, autoimmunity, sepsis). Brown adipocytes can likewise produce inflammatory cytokines, via activation of their thermogenic programming, however this area is nascent and existing literature remains limited47. In addition to classically quantified inflammatory mediators (e.g., TNF and IL-6) adipocytes, like macrophages and DCs, express other inflammatory mediators that are not traditionally reported outside of the field of immunology, including B cell activating factor

(BAFF) and A proliferation-inducing ligand (APRIL)29,30. Both BAFF and APRIL modify B cell maturation and survival48,49 and have been correlated with promoting inflammation in multiple autoimmune diseases48. B cells are well established to infiltrate WAT in the context of obesity 50 and to modulate obesity pathogenesis 51. Thus, whether adipocytes, via BAFF and APRIL production, can modify B cell status within the WAT and impact disease pathogenesis remains unknown. In contrast, adipocytes also express all three known receptors of BAFF and APRIL including BAFF receptor (BAFF-R), transmembrane activator and CAML interactor (TACI), and B cell maturation antigen (BCMA) 29. Whether

BAFF and APRIL production by AT infiltrating macrophages or DCs, and subsequent sensing by adipocytes can alter adipocyte function and/or the crosstalk among myeloid cells, adipocytes and adaptive immune cells in AT is an intriguing inquiry. The importance of adipocyte and immune cell crosstalk likely extends beyond cytokines, as white adipocytes produce multiple chemokines (e.g., CCL2, CCL5, CXCL8, CXCL10) 26,32 and brown adipocytes can produce CXCL1452. Whether this crosstalk is more expansive and includes adipocyte production of other well-established immune cell recruiting

8 chemokines (e.g., CCL3, CXCL1, CXCL9) or immune cell skewing cytokines (e.g., TGF- b, IFN-g, IL-1b, IL-4, IL-10, IL-23) is understudied. The impact of such knowledge may shed light on how adipocytes affect immune cell polarization (e.g., M1/M2 macrophages,

Th1, Th2, Th17, regulatory T cells [Tregs], B-1 and B-2 cells) and subsequent disease pathogenesis. Conversely, whether adipocytes can sense pro- and anti-inflammatory mediators and subsequently are themselves skewed/polarized towards a “pro- or anti- inflammatory” phenotype is unknown.

The underpinning mechanisms with the capacity to regulate and uncover adipocyte-intrinsic inflammatory potential are underdefined. Epigenetic modifications alter cytokine production in immune cells53. Adipocyte epigenome is an established contributor to their thermogenic programming54 hence, the epigenome may similarly impact adipocyte inflammatory programming. Further, adipocytes are highly metabolic cells and cellular metabolic programming in immune cells is intricately tied their inflammatory vigor55,56. As proinflammatory cytokines (e.g., TNF, IL-6, Type I IFNs, IL-2) are known to modulate core immune cell metabolism, if they similarly regulate adipocyte metabolic capacity remains underdefined. Conversely, if alteration of adipocyte metabolic capacity is tied to adipocyte “immune-like” potential is unknown.

Adipocyte expression of innate immune receptors and production of proinflammatory mediators insinuate the possibility that adipocytes converse with adaptive immune cells. Recent studies have shed some light into direct crosstalk between adipocytes and adaptive immunity. Initial microarray analyses of primary adipocytes uncovered increased expression of major histocompatibility complex II (MHCII) and genes involved with antigen processing and presentation (e.g., CIITA, CD80, CD86,

CD40)34,36,57. Closer examination revealed that antigen-specific presenting adipocytes activated CD4+ T cells, in a direct manner, demonstrating the functionality of this antigen processing machinery. Notably, this capacity to present antigen and activate CD4+ T cells was largely attributed to large adipocytes36. As adipocyte size is a large determinant of metabolic health58,59, with larger adipocytes typically an indicator of metabolic derangements, these findings may suggest that adipocytes may be an overlooked contributor to MHC-driven processes in metabolic diseases. Recent examination of adipocyte-specific deletion of MHCII revealed dampened obesity-associated AT-driven

IFNg production and an enhanced abundance of Tregs within AT3. While lack of adipocyte-specific MHCII preserved insulin sensitivity and glucose metabolism in the context of obesity, CD80/86 and CD40 are proposed to play beneficial roles in maintenance of insulin resistance33,57. Hence studies to formally parse apart the interplay between MHCII and co-stimulation, and their underpinning mechanisms, in adipocytes are needed. Additionally, broader understanding of whether adipocytes express MHC class I and can activate CD8+ T cells is unknown and should be studied. Adipocytes also highly express CD1d and directly present lipid antigens to activate invariant NKT (iNKT) cells37. AT iNKT numbers and activity have been reported to decline in obesity60. iNKT deficient mice exhibit insulin resistance, adipocyte hypertrophy, suggesting an interplay between adipocytes and iNKT cells37. Future investigations focused on the interplay between adipocytes and iNKT cells in disease pathogenesis are warranted.

Collectively, published findings suggest that adipocytes may contribute to the crosstalk with immune cells in a manner that is potentially either beneficial or detrimental to disease pathogenesis – the outcome being dependent on disease setting. Thus, it is

10 conceivable that adipocyte function and inflammatory capacity may modulate pathogenesis of metabolic, cancer, autoimmune and infectious diseases. Published evidence of how adipocytes may impact various diseases are discussed below.

INFLAMMATION AND ALTERATION OF ADIPOCYTE FUNCTION

Core observations that patients with increased circulating proinflammatory cytokines (e.g., TNF, IL-6) display augmented energy expenditure and mobilization of lipids, suggests a connection between inflammation and AT function61,62. Additional characterizations have demonstrated that inflammatory cytokines (TNF63-67, IL-1b64,68, IL-

669,70, IL-471, IL-17a72, IFNa64, IFNg64) are capable of inducing lipolysis in 3T3-L1 cells, a traditionally utilized adipocyte-like cell line23. Although adipocyte-specific deletion of a rate limiting enzyme, adipose triglyceride lipase (ATGL), of lipolysis impacted obesity-driven hepatic immune cell infiltration and activation, it was insufficient to dampen AT inflammation73. In contrast, deletion of a separate rate limiting enzyme of lipolysis, hormone sensitive lipase (HSL), resulted in enhanced AT inflammation74. Hence more extensive investigations are necessary to define the relevance of cytokine-mediated lipolysis in vivo towards systemic inflammation and disease pathology. Additionally, the source of the cytokines (adipocytes and/or immune cells) and whether these proinflammatory cytokines drive lipolysis in all types of primary adipocytes (e.g., white, brown, beige, pink) or if they regulate contrasting functions is unknown.

Adipocyte sensing of cytokines modifies adipocyte energy storage machinery. In addition to promoting lipolysis, TNF inhibits FA uptake, via downregulation of FA transport protein (FATP) and fatty acid translocase (FAT) and lipogenesis in vitro75,76. Although,

11 these collective findings suggest proinflammatory cytokines may drive detrimental effects in WAT, recent evidence has emerged to suggest that inflammation may also play vital beneficial roles in adipocyte health. Specifically, lack of AT TNF sensing in vivo prohibits accumulation of lipids into WAT, leading to exacerbated metabolic dysfunction and ectopic lipid accumulation4. Mirroring these findings, lack of adipocyte-specific TLR4 sensing dysregulates WAT expansion and exacerbates metabolic disease77. Despite this, the contribution of other TLRs and inflammatory mediators (e.g., IL-6, IL-17, IL-1b, BAFF,

APRIL) in adipocytes to obesity-associated inflammation remains underdefined with further investigations required to mechanistically comprehend adipocyte-specific contributions to WAT remodeling.

Well studied signaling pathways downstream of PAMP and cytokine recognition in immune cells including NF-kB78,79 and Jak/STATs (e.g., Jak1/2/3, Tyk2,

STAT1/3/5A/5B)80 appear to play a similar role in adipocytes. Ability of these pathways to modulate adipocyte lipolysis and adipogenesis and has been previously reviewed80. In addition, NF-kB and Jak2 may also play a role in modulation of UCP-1 dependent thermogenesis81-83, however tissue specific genetic investigations are necessary to provide greater clarity on its impact on brown adipocytes. One potential unifying immune mediator is the NF-kB regulated transcription factor interferon regulatory factor 4 (IRF4), an established modulator of lymphocyte development and function84 that can beneficially modify both white adipocyte lipolysis and brown adipocyte adaptive thermogenesis85. In addition, Jak/STAT are critical mediators of interferon α/β receptor (IFNAR) signaling, a ubiquitously expressed receptor that regulates both innate and adaptive inflammatory potential. Notably, interferon (IFN) β impacts thermogenic capacity of brown adipocytes86.

Whether type I IFN is sensed by white adipocytes and if this uncovers their inflammatory potential, akin to immune cell counterparts87-89, is unknown and should be formally examined. Given that adipocytes have an “immune-like” capacity and express similar signaling pathways to immune cells, it is plausible that the regulatory mechanisms (e.g., epigenetic, transcriptional, metabolic) likewise overlap between these cells.

The existence of beige adipocytes was hinted at over three decades ago, wherein brown adipocytes were observed in WAT90, something that was enhanced under cold stress91. While mechanisms regulating development of the beige adipocyte lineage and modulation of their function is still heavily investigated, links have emerged suggesting immune mediators and inflammation may modulate beige adipocyte function.

Sympathetic stimulation, via catecholamines, drives beige adipogenesis7. A suggestion that sensing of IL-4 by AT alternatively activated macrophages upregulates tyrosine hydroxylase (TH; rate limiting enzyme of norepinephrine [NE] synthesis) and contributes to a local NE production and subsequent adipocyte beiging was proposed92,93 and elicited significant interest in the field. However, a detailed follow-up study revealed that bone- marrow derived macrophages do not produce NE in response to IL-4 nor do AT resident macrophages express TH94. Further, neither hematopoietic-specific TH deletion nor chronic exposure to IL-4, in WT or IL-4ra-/- mice was sufficient to modify energy expenditure94. These latter findings clearly suggest existence of alternate mechanisms linking IL-4-driven macrophage contribution to beiging — one being the capacity for macrophages to capture and release catecholamines in an IL-4 regulated manner.

In contrast to IL-4, autocrine and paracrine TGFb signaling has been demonstrated to constitutively suppress beige adipogenesis 95. Activation of Jak/STAT3 pathway, via

IL-6 and IL-11, inhibits TGFb suppression of beiging 95. It is posited that sympathetic stimulation augments adipocyte lipolysis which drives a transient elevation in IL-6/IL-11 to promote beiging of adipocyte progenitors in a paracrine fashion 95. However, whether the source of these cytokines is from adipocytes or resident AT immune cells remains undefined. Further, in states of acute (e.g., sepsis/infection) or chronic inflammation (e.g., obesity, autoimmunity), whether this pathway of beiging becomes tolerized or altered is not well understood.

While cytokines and inflammation may have indirect means to drive adipocyte beiging/browning, recent evidence points to the possibility of direct means wherein immune cells can directly regulate adipocyte beiging. Mature adipocytes isolated directly from AT express vascular adhesion protein 1 (VCAM-1) 96. Expression of VCAM-1, which is enhanced in obesity, facilitates direct interaction with a4-integrin on macrophages driving a reciprocal modulation of macrophage activation and downregulation of UCP-1 expression in adipocytes 96. Lack of a4-integrin promoted beiging and improved obesity- associated metabolic derangements 96, suggest that this direct interaction between macrophages and adipocytes facilitates a key component to beiging. However, whether

UCP-1 independent thermogenesis is likewise modulated by this interaction remains unknown. Combined, existing literature suggests that adipocyte (white, brown, beige) sensing of inflammatory mediators and interaction with immune cells can modify adipocyte function. However, whether adipocyte to adipocyte communication, via inflammatory mediators, released lipids, can also alter adipocyte “immune-like” potential and function is poorly understood and should be studied.

ADIPOCYTE-INFLAMMATION AXIS IN DISEASE PATHOGENESIS

Among various diseases, AT contribution to obesity pathogenesis and obesity-associated inflammation has naturally been the most well examined6,12,59. Obesity-associated chronic low-grade inflammation is central to virtually all downstream derangements including type 2 diabetes (T2D), atherosclerotic cardiovascular disease (CVD), nonalcoholic fatty liver disease (NAFLD), Alzheimer’s disease, diverse cancers and increased complications from infectious sequelae97,98. Obesity-driven enhancement of inflammatory mediators, including TNF, IL-6, IL-17, are well defined critical mediators of obesity- associated NAFLD 99-103 and CVD 104-106 pathogenesis (Figure 3). Despite the intuitive sense that adipocytes may contribute to obesity-associated inflammation and downstream derangements, given their “immune-like” potential and sensing of inflammatory mediators to modify functional capacity, whether and how obesity exacerbates adipocyte inflammatory capacity, and whether adipocyte-intrinsic production of these proinflammatory mediators drives disease pathogenesis in obesity is not well understood. One hint interlinking obesity with adipocyte inflammation stems from evidence that lack of adipocyte-specific TLR4 sensing, in the context of obesogenic-diet feeding, exhibits both acute (e.g., protection from insulin resistance) and chronic-term effects (e.g., lack of AT remodeling, ectopic lipid deposition, exacerbation metabolic disease). Adipocyte-specific TLR4 deletion led to decreased TLR4 expression in other tissues, including the liver and peritoneal macrophages 77. These findings reaffirm that adipocytes are critical and complex players in global metabolic health. However, additional mechanistic interrogations focused on the contribution of adipocyte-specific inflammation to endocrine regulation of systemic tissues is warranted.

While the mechanisms that uncover adipocyte inflammatory capacity remain unknown, undifferentiated 3T3-L1 cells can produce type I IFN 107, a regulator of innate and adaptive immune responses. Whether primary adipocytes can produce type I IFNs and if sensing of type I IFNs can modulate adipocyte inflammatory potential should be elucidated and further investigated. Extensive literature underscores the contribution of the type I IFN/IFNAR axis as a promoter of metabolic dysfunction (e.g., regulation of intrahepatic CD8+ T cell pathogenicity 108, interferon regulatory factors 109-111, suppression of thermogenesis 86, exacerbation of insulin resistance and hepatic dysfunction 112-115). In contrast, a recent study demonstrated that adipocyte- deletion of IFNAR, via use of fatty acid binding protein 4 (FABP4) cre, augmented obesogenic-diet induced metabolic disease 116. However, FABP4 expression is not restricted to adipocytes (gene expression in macrophages, endothelial cells, osteogenic cells, ganglion, adrenal medulla and liver could be affected 117,118) thus conclusions driving the specific contribution of adipocyte

IFNAR signaling to metabolic disease require additional interrogations—something that could be addressed using a more specific adipocyte-targeted in vivo deletion using

Adipoqcre mice 118.

Studies of AT/adipocyte-specific contribution in other pathologies remain limited including cancer, autoimmunity and infections. Obesity is a major risk factor for the development and pathogenesis of 13 different forms of cancer including breast, endometrial, liver, pancreatic, colorectal and ovarian8. Increased systemic FAs and inflammatory cytokines are posited mechanisms fueling cancer pathogenesis. Obesity- driven AT milieu has been linked to tumor pathogenesis119. Similarly, both adipocyte paracrine and endocrine effects are suggested to enhance tumor pathogenesis120.

Adipocyte release of FA, via lipolysis, may represent a potential fuel source to tumorigenic cells. In turn, FA can enhance b-oxidation, a modulator of immune cell inflammatory vigor121, within oncogenic cells and promote their migration and invasiveness122,123. Thus, whether cytokine-driven adipocyte lipolysis plays a role in cancer pathogenesis should be examined. Further, if adipocytes-intrinsic production of proinflammatory cytokines (e.g.,

TNF, IL-6) and/or crosstalk with immune cells (e.g., Tregs) directly impact proliferation, survival, invasiveness, and metastasis of tumors should be examined.

Like obesity, there is a sharp rise in development of autoimmune diseases in developed countries. Autoimmune diseases (e.g., psoriasis, rheumatoid arthritis [RA]) are associated with AT inflammation and increased immune cell influx into the AT124,125. B cells within human AT can produce autoimmune antibodies32. These collective events in autoimmunity are energy exhaustive processes. Cytokines that are augmented in autoimmunity can likewise facilitate adipocyte lipid handling (e.g., TNF) and potentially provide a source of fuel to the autoimmune inflammatory flame. Given that adipocytes can produce inflammatory mediators known to modulate pathology of autoimmune diseases (e.g., TNF, IL-17, BAFF, APRIL and IL-4), the contribution of adipocyte-specific inflammation to autoimmunity should be defined. Hence, the direct or indirect impact of adipocyte-specific contribution to T- and B cell-driven autoimmunity should be examined.

Accumulation of AT, through obesity, is linked with greater complications, morbidity and mortality to infections98. Notably, these studies have exclusively examined

WAT/white adipocytes. For instance, a close relationship between AT/adipocytes and HIV has been proposed. HIV infected individuals exhibit T cell infiltration into AT, lipodystrophy and metabolic dysfunction5. Co-culturing of HIV infected T cells with adipocytes enhanced

17 detection of HIV replication subunits126. Adipocytes express the necessary receptors for

HIV entry (e.g., CD4, CCR5 and CXCR4)127 and thus it is plausible that adipocytes are an important long-term reservoir of HIV that mediates relapse of infection126. Whether HIV itself can trigger changes in adipocyte-intrinsic function needs to be further elucidated.

In addition to HIV, various bacterial (e.g., mycobacterium tuberculosis [Mtb]), viral

(e.g., cytomegalovirus, influenza A, respiratory syncytial virus, adenovirus) and parasitic

(e.g., Trypanosoma cruzi) pathogens can infect adipocytes5,128,129. Transfer of WAT from

Mtb infected mice to an uninfected host contributed to Mtb dissemination128. One intriguing conjecture, given that adipocytes are a rich energy source, is that evolutionarily it would have been energetically advantageous for pathogens to develop the ability to reside within adipocytes. Whether pathogens can likewise infect other subsets of adipocytes (brown, beige, pink) and can trigger an adipocyte’s “immune-like” potential is unknown and should be explored. Further, if adipocyte-specific inflammation contributes beneficially (e.g., immune cell polarization, pathogen clearance) or detrimentally (e.g., bystander host damage, dampened immune cell function) in the context of infections should be formally defined. Lastly, as an extension to apparent correlations, examination of adipocyte contribution to other obesity-impacted health complications (e.g., neurological diseases, aging, reproductive fitness) is lacking and would be of significant future interest.

CONCLUSIONS

Overall, this review highlights the necessity to consider and evaluate the contribution of adipocyte-specific inflammation within the enclave of AT inflammatory capacity. The

18 reciprocal nature between adipocytes and inflammation is grossly underappreciated. The expansive knowledge gain of mechanisms regulating immune cell inflammation could be harnessed and in turn explored in adipocytes. In addition to greater definition of crosstalk between adipocytes and immune cells, whether adipocyte to adipocyte communication occurs (e.g., white adipocyte induction of brown/beige adipocyte thermogenesis), via cytokines, is unknown and would be an exciting area to explore. Definition of these mechanisms regulating adipocyte-intrinsic inflammatory capabilities and contribution to disease pathogenesis (Figure 4) could uncover previously unappreciated therapeutic avenues. Notably, existing therapeutics targeting inflammatory pathways in immune cells could be readily repurposed and utilized to target adipocytes and adipocyte inflammation.

ACKNOWLEDGEMENTS

This study was supported, in part, by CCHMC Pediatric Diabetes and Obesity Center (to

S.D.), NIH R01DK099222 (to S.D); NIH T32AI118697 and T32GM063483-14 (associated with C.C.C.).

AUTHOR CONTRIBUTIONS

C.C.C., M.S.M.A.D., P.C.A., J.S-G. S.D. participated in the conception of this review and wrote the manuscript.

COMPETING FINANCIAL INTERESTS

The authors declare no competing financial interests

Figure 1. Adipocyte subsets. Adipocytes are divided in various subtypes: white adipocytes (energy storing); brown adipocytes (thermogenesis); beige/brite adipocytes (inducible energy expending); pink adipocytes (pregnancy/lactation associated). White adipocytes are characterized by large lipid droplets and are known for their energy storing capacity during obesity. Brown adipocytes are characterized by numerous mitochondria. They originate from a separate preadipocyte population. Mitochondrial number and function within brown adipocytes support their thermogenic capacity and maintenance of body temperature. Beige adipocytes arise within white adipose tissue (WAT) and are characterized by unique thermogenic capacity. Induction of beige adipocytes is driven by environmental cues (e.g., cold stress, exercise, b3-adrenergic stimulation). The developmental origins of beige adipocytes is still under investigation. Pink adipocytes are proposed to transdifferentiate from cells within subcutaneous AT. The precise function and developmental origins of pink adipocytes remains underdefined.

Figure 2. Adipocyte-centric inflammation and crosstalk with immune cells. Adipocytes possess an “immune-like” potential including: (a) expression of innate immune receptors (e.g., Toll-like receptors [TLRs], NOD-like receptors [NLR], RIG-I receptors [RLR]); (b) cytokine (e.g., TNF, IL1β, IL-6, BAFF, APRIL) and chemokine (CCL2, CXCL8, CXCL1, CXCL10) production; and (c) antigen presentation (e.g., MHC-II, CD80, CD40, CD1d, CIITA). Adipocytes sensing of cytokines alters their function (e.g., lipolysis, lipogenesis, UCP-1 thermogenesis). Cumulatively, these adipocyte inflammatory features suggest potential crosstalk with immune cells (in AT or circulation) that may directly impact adipocyte function, immune cell function, and/or disease pathogenesis.

Normal WAT Obese WAT

Macrophage Homeostatic: Dysregulated: • Cellular milieu • Cytokines • Cytokines • Immune cell infiltration Dendritic cells Normal BAT Obese BAT T cells B cells

iNKTcells

Homeostatic: Dysregulated: • Cellular milieu • Immune cell infiltration • Cytokines

Figure 3. Obesity-driven AT inflammatory plasticity. White and brown AT in a lean state consists of a diverse cellular milieu (e.g., adipocytes, macrophages, T cells, B cells) and capacity to produce cytokines. Pathological expansion of AT in obesity promotes dysregulation of cytokine production (e.g., TNF, IL-6, IL-1, IFNγ), enhances immune cell infiltration/polarization/activation and exacerbates the AT inflammatory milieu. While contribution of AT to obesity-associated inflammation is well investigated, adipocyte- specific contribution to inflammation is understudied and remains a critical gap in knowledge.

Cytokine receptors Obesity associated Innate immune receptors (e.g. metabolic diseases TLRs, NLRP3, (e.g. T2D, CVD, RLPs) Cellular Metabolism NAFLD) STAT (lipolysis, lipogenesis) Antigen presentation JAK ? glycolysis/fatty \acid oxidation NF-!B e.g. IL6, TNF, BAFF Cytokine production

Epigenetic changes Cytokine TLRs ? Infectious receptors diseases ?

? ? NF -!B JAK ? STAT Cancer NOD

Cytokine Autoimmunity ? production

Figure 4. Adipocyte contribution to disease. Adipocytes have “immune-like” capabilities that, in part, overlap with traditionally investigated immune cells. White adipocyte-specific inflammatory capacity can impact obesity-associated metabolic diseases. The contributions of other subsets of adipocytes (e.g., brown) to disease should be formally examined. Relevance of the interconnection in adipocytes between signaling pathways (e.g., NF-kB, Jak/STAT), core metabolism (e.g., glycolysis, b-oxidation) and epigenetic programming to promote adipocyte-intrinsic inflammatory capacity and contribute to disease pathogenesis warrants future investigation. Additionally, if adipocyte crosstalk with immune cells likewise contributes to diseases including autoimmunity, cancer and infections are a natural extension to current investigations.

REFERENCES

1. Ghaben, A.L. & Scherer, P.E. Adipogenesis and metabolic health. Nat Rev Mol Cell Biol (2019). 2. Sarantopoulos, C.N., et al. Elucidating the Preadipocyte and Its Role in Adipocyte Formation: a Comprehensive Review. Stem Cell Rev 14, 27-42 (2018). 3. Deng, T., et al. Adipocyte adaptive immunity mediates diet-induced adipose inflammation and insulin resistance by decreasing adipose Treg cells. Nature Communications 8(2017). 4. Wernstedt Asterholm, I., et al. Adipocyte inflammation is essential for healthy adipose tissue expansion and remodeling. Cell Metab 20, 103-118 (2014). 5. Desruisseaux, M.S., Nagajyothi, Trujillo, M.E., Tanowitz, H.B. & Scherer, P.E. Adipocyte, adipose tissue, and infectious disease. Infect Immun 75, 1066-1078 (2007). 6. Grant, R.W. & Dixit, V.D. Adipose tissue as an immunological organ. Obesity (Silver Spring) 23, 512-518 (2015). 7. Kajimura, S., Spiegelman, B.M. & Seale, P. Brown and Beige Fat: Physiological Roles beyond Heat Generation. Cell Metab 22, 546-559 (2015). 8. Park, J., Morley, T.S., Kim, M., Clegg, D.J. & Scherer, P.E. Obesity and cancer-- mechanisms underlying tumour progression and recurrence. Nat Rev Endocrinol 10, 455-465 (2014). 9. Enerback, S. Adipose tissue metabolism in 2012: Adipose tissue plasticity and new therapeutic targets. Nat Rev Endocrinol 9, 69-70 (2013). 10. Ouchi, N., Parker, J.L., Lugus, J.J. & Walsh, K. Adipokines in inflammation and metabolic disease. Nat Rev Immunol 11, 85-97 (2011). 11. Villarroya, F., Cereijo, R., Villarroya, J., Gavalda-Navarro, A. & Giralt, M. Toward an Understanding of How Immune Cells Control Brown and Beige Adipobiology. Cell Metab 27, 954-961 (2018). 12. Ghigliotti, G., et al. Adipose tissue immune response: novel triggers and consequences for chronic inflammatory conditions. Inflammation 37, 1337-1353 (2014). 13. Huh, J.Y., Park, Y.J., Ham, M. & Kim, J.B. Crosstalk between adipocytes and immune cells in adipose tissue inflammation and metabolic dysregulation in obesity. Mol Cells 37, 365-371 (2014). 14. Herrero, L., Shapiro, H., Nayer, A., Lee, J. & Shoelson, S.E. Inflammation and adipose tissue macrophages in lipodystrophic mice. Proc Natl Acad Sci U S A 107, 240-245 (2010). 15. Fitzgibbons, T.P., et al. Similarity of mouse perivascular and brown adipose tissues and their resistance to diet-induced inflammation. Am J Physiol Heart Circ Physiol 301, H1425-1437 (2011). 16. Sanchez-Gurmaches, J., Hung, C.M. & Guertin, D.A. Emerging Complexities in Adipocyte Origins and Identity. Trends Cell Biol 26, 313-326 (2016). 17. Tan, C.Y. & Vidal-Puig, A. Adipose tissue expandability: the metabolic problems of obesity may arise from the inability to become more obese. Biochem Soc Trans 36, 935-940 (2008).

18. Lafontan, M. & Langin, D. Lipolysis and lipid mobilization in human adipose tissue. Prog Lipid Res 48, 275-297 (2009). 19. Frayn, K.N. Adipose tissue as a buffer for daily lipid flux. Diabetologia 45, 1201- 1210 (2002). 20. Ikeda, K., Maretich, P. & Kajimura, S. The Common and Distinct Features of Brown and Beige Adipocytes. Trends Endocrinol Metab 29, 191-200 (2018). 21. Petrovic, N., et al. Chronic peroxisome proliferator-activated receptor gamma (PPARgamma) activation of epididymally derived white adipocyte cultures reveals a population of thermogenically competent, UCP1-containing adipocytes molecularly distinct from classic brown adipocytes. J Biol Chem 285, 7153-7164 (2010). 22. Cinti, S. Pink Adipocytes. Trends Endocrinol Metab 29, 651-666 (2018). 23. Ruiz-Ojeda, F.J., Ruperez, A.I., Gomez-Llorente, C., Gil, A. & Aguilera, C.M. Cell Models and Their Application for Studying Adipogenic Differentiation in Relation to Obesity: A Review. Int J Mol Sci 17(2016). 24. Ballak, D.B., et al. TLR-3 is present in human adipocytes, but its signalling is not required for obesity-induced inflammation in adipose tissue in vivo. PLoS One 10, e0123152 (2015). 25. Batra, A., et al. Leptin-dependent toll-like receptor expression and responsiveness in preadipocytes and adipocytes. Am J Pathol 170, 1931-1941 (2007). 26. Kopp, A., et al. Innate immunity and adipocyte function: ligand-specific activation of multiple Toll-like receptors modulates cytokine, adipokine, and chemokine secretion in adipocytes. Obesity (Silver Spring) 17, 648-656 (2009). 27. Lin, Y., et al. The lipopolysaccharide-activated toll-like receptor (TLR)-4 induces synthesis of the closely related receptor TLR-2 in adipocytes. J Biol Chem 275, 24255-24263 (2000). 28. Pietsch, J., et al. Toll-like receptor expression and response to specific stimulation in adipocytes and preadipocytes: on the role of fat in inflammation. Ann N Y Acad Sci 1072, 407-409 (2006). 29. Alexaki, V.I., et al. Adipocytes as immune cells: differential expression of TWEAK, BAFF, and APRIL and their receptors (Fn14, BAFF-R, TACI, and BCMA) at different stages of normal and pathological adipose tissue development. J Immunol 183, 5948-5956 (2009). 30. Muller, N., et al. B Lymphocyte Stimulator (BLyS) is expressed in human adipocytes in vivo and is related to obesity but not to insulin resistance. PLoS One 9, e94282 (2014). 31. Sewter, C.P., Digby, J.E., Blows, F., Prins, J. & O'Rahilly, S. Regulation of tumour necrosis factor-alpha release from human adipose tissue in vitro. J Endocrinol 163, 33-38 (1999). 32. Frasca, D., Diaz, A., Romero, M., Thaller, S. & Blomberg, B.B. Secretion of autoimmune antibodies in the human subcutaneous adipose tissue. PLoS One 13, e0197472 (2018). 33. Guo, C.A., et al. CD40 deficiency in mice exacerbates obesity-induced adipose tissue inflammation, hepatic steatosis, and insulin resistance. Am J Physiol Endocrinol Metab 304, E951-963 (2013).

34. Deng, T., et al. Class II major histocompatibility complex plays an essential role in obesity-induced adipose inflammation. Cell Metab 17, 411-422 (2013). 35. Poggi, M., et al. The inflammatory receptor CD40 is expressed on human adipocytes: contribution to crosstalk between lymphocytes and adipocytes. Diabetologia 52, 1152-1163 (2009). 36. Xiao, L., et al. Large adipocytes function as antigen-presenting cells to activate CD4(+) T cells via upregulating MHCII in obesity. Int J Obes (Lond) 40, 112-120 (2016). 37. Huh, J.Y., et al. A novel function of adipocytes in lipid antigen presentation to iNKT cells. Mol Cell Biol 33, 328-339 (2013). 38. O'Neill, L.A., Golenbock, D. & Bowie, A.G. The history of Toll-like receptors - redefining innate immunity. Nat Rev Immunol 13, 453-460 (2013). 39. Bae, J., et al. Activation of pattern recognition receptors in brown adipocytes induces inflammation and suppresses uncoupling protein 1 expression and mitochondrial respiration. Am J Physiol Cell Physiol 306, C918-930 (2014). 40. Kanneganti, T.D. Central roles of NLRs and inflammasomes in viral infection. Nat Rev Immunol 10, 688-698 (2010). 41. Vandanmagsar, B., et al. The NLRP3 inflammasome instigates obesity-induced inflammation and insulin resistance. Nat Med 17, 179-188 (2011). 42. Zhou, Y.J., et al. Increased NOD1, but not NOD2, activity in subcutaneous adipose tissue from patients with metabolic syndrome. Obesity (Silver Spring) 23, 1394- 1400 (2015). 43. Zhou, Y.J., Zhou, H., Li, Y. & Song, Y.L. NOD1 activation induces innate immune responses and insulin resistance in human adipocytes. Diabetes Metab 38, 538- 543 (2012). 44. Yu, L., et al. Pattern recognition receptor-initiated innate antiviral response in mouse adipose cells. Immunol Cell Biol 92, 105-115 (2014). 45. Cani, P.D., et al. Metabolic endotoxemia initiates obesity and insulin resistance. Diabetes 56, 1761-1772 (2007). 46. Nishimoto, S., et al. Obesity-induced DNA released from adipocytes stimulates chronic adipose tissue inflammation and insulin resistance. Sci Adv 2, e1501332 (2016). 47. Mauer, J., et al. Signaling by IL-6 promotes alternative activation of macrophages to limit endotoxemia and obesity-associated resistance to insulin. Nat Immunol 15, 423-430 (2014). 48. Mackay, F., Schneider, P., Rennert, P. & Browning, J. BAFF AND APRIL: a tutorial on B cell survival. Annu Rev Immunol 21, 231-264 (2003). 49. Schneider, P., et al. BAFF, a novel ligand of the tumor necrosis factor family, stimulates B cell growth. J Exp Med 189, 1747-1756 (1999). 50. Winer, D.A., et al. B cells promote insulin resistance through modulation of T cells and production of pathogenic IgG antibodies. Nat Med 17, 610-617 (2011). 51. DeFuria, J., et al. B cells promote inflammation in obesity and type 2 diabetes through regulation of T-cell function and an inflammatory cytokine profile. Proc Natl Acad Sci U S A 110, 5133-5138 (2013).

52. Cereijo, R., et al. CXCL14, a Brown Adipokine that Mediates Brown-Fat-to- Macrophage Communication in Thermogenic Adaptation. Cell Metab 28, 750-763 e756 (2018). 53. Phan, A.T., Goldrath, A.W. & Glass, C.K. Metabolic and Epigenetic Coordination of T Cell and Macrophage Immunity. Immunity 46, 714-729 (2017). 54. Peng, X., Zhang, Q., Liao, C., Han, W. & Xu, F. Epigenomic Control of Thermogenic Adipocyte Differentiation and Function. Int J Mol Sci 19(2018). 55. O'Neill, L.A., Kishton, R.J. & Rathmell, J. A guide to immunometabolism for immunologists. Nat Rev Immunol 16, 553-565 (2016). 56. Pearce, E.J. & Pearce, E.L. Immunometabolism in 2017: Driving immunity: all roads lead to metabolism. Nat Rev Immunol 18, 81-82 (2018). 57. Zhong, J., et al. T-cell costimulation protects obesity-induced adipose inflammation and insulin resistance. Diabetes 63, 1289-1302 (2014). 58. McLaughlin, T., et al. Adipose Cell Size and Regional Fat Deposition as Predictors of Metabolic Response to Overfeeding in Insulin-Resistant and Insulin-Sensitive Humans. Diabetes 65, 1245-1254 (2016). 59. Tandon, P., Wafer, R. & Minchin, J.E.N. Adipose morphology and metabolic disease. J Exp Biol 221(2018). 60. Schipper, H.S., et al. Natural killer T cells in adipose tissue prevent insulin resistance. J Clin Invest 122, 3343-3354 (2012). 61. Spitzer, J.J., Bagby, G.J., Meszaros, K. & Lang, C.H. Alterations in lipid and carbohydrate metabolism in sepsis. JPEN J Parenter Enteral Nutr 12, 53S-58S (1988). 62. Kamimura, M.A., et al. Serum and cellular interleukin-6 in haemodialysis patients: relationship with energy expenditure. Nephrol Dial Transplant 22, 839-844 (2007). 63. Green, A., Rumberger, J.M., Stuart, C.A. & Ruhoff, M.S. Stimulation of lipolysis by tumor necrosis factor-alpha in 3T3-L1 adipocytes is glucose dependent: implications for long-term regulation of lipolysis. Diabetes 53, 74-81 (2004). 64. Feingold, K.R., Doerrler, W., Dinarello, C.A., Fiers, W. & Grunfeld, C. Stimulation of lipolysis in cultured fat cells by tumor necrosis factor, interleukin-1, and the interferons is blocked by inhibition of prostaglandin synthesis. Endocrinology 130, 10-16 (1992). 65. Gasic, S., Tian, B. & Green, A. Tumor necrosis factor alpha stimulates lipolysis in adipocytes by decreasing Gi protein concentrations. J Biol Chem 274, 6770-6775 (1999). 66. Zhang, H.H., Halbleib, M., Ahmad, F., Manganiello, V.C. & Greenberg, A.S. Tumor necrosis factor-alpha stimulates lipolysis in differentiated human adipocytes through activation of extracellular signal-related kinase and elevation of intracellular cAMP. Diabetes 51, 2929-2935 (2002). 67. Laurencikiene, J., et al. NF-kappaB is important for TNF-alpha-induced lipolysis in human adipocytes. J Lipid Res 48, 1069-1077 (2007). 68. Price, S.R., Mizel, S.B. & Pekala, P.H. Regulation of lipoprotein lipase synthesis and 3T3-L1 adipocyte metabolism by recombinant interleukin 1. Biochim Biophys Acta 889, 374-381 (1986). 69. Zhang, W., et al. Adipocyte lipolysis-stimulated interleukin-6 production requires sphingosine kinase 1 activity. J Biol Chem 289, 32178-32185 (2014).

70. Ji, C., et al. IL-6 induces lipolysis and mitochondrial dysfunction, but does not affect insulin-mediated glucose transport in 3T3-L1 adipocytes. J Bioenerg Biomembr 43, 367-375 (2011). 71. Tsao, C.H., Shiau, M.Y., Chuang, P.H., Chang, Y.H. & Hwang, J. Interleukin-4 regulates lipid metabolism by inhibiting adipogenesis and promoting lipolysis. J Lipid Res 55, 385-397 (2014). 72. Shin, J.H., Shin, D.W. & Noh, M. Interleukin-17A inhibits adipocyte differentiation in human mesenchymal stem cells and regulates pro-inflammatory responses in adipocytes. Biochem Pharmacol 77, 1835-1844 (2009). 73. Schoiswohl, G., et al. Impact of Reduced ATGL-Mediated Adipocyte Lipolysis on Obesity-Associated Insulin Resistance and Inflammation in Male Mice. Endocrinology 156, 3610-3624 (2015). 74. Xia, B., et al. Adipose tissue deficiency of hormone-sensitive lipase causes fatty liver in mice. PLoS Genet 13, e1007110 (2017). 75. Memon, R.A., Feingold, K.R., Moser, A.H., Fuller, J. & Grunfeld, C. Regulation of fatty acid transport protein and fatty acid translocase mRNA levels by endotoxin and cytokines. Am J Physiol 274, E210-217 (1998). 76. Ruan, H., Pownall, H.J. & Lodish, H.F. Troglitazone antagonizes tumor necrosis factor-alpha-induced reprogramming of adipocyte gene expression by inhibiting the transcriptional regulatory functions of NF-kappaB. J Biol Chem 278, 28181- 28192 (2003). 77. Tao, C., et al. Short-Term Versus Long-Term Effects of Adipocyte Toll-Like Receptor 4 Activation on Insulin Resistance in Male Mice. Endocrinology 158, 1260-1270 (2017). 78. Berg, A.H., Lin, Y., Lisanti, M.P. & Scherer, P.E. Adipocyte differentiation induces dynamic changes in NF-kappaB expression and activity. Am J Physiol Endocrinol Metab 287, E1178-1188 (2004). 79. Weidemann, A., et al. Classical and alternative NF-kappaB signaling cooperate in regulating adipocyte differentiation and function. Int J Obes (Lond) 40, 452-459 (2016). 80. Xu, D., Yin, C., Wang, S. & Xiao, Y. JAK-STAT in lipid metabolism of adipocytes. JAKSTAT 2, e27203 (2013). 81. Chiang, S.H., et al. The protein kinase IKKepsilon regulates energy balance in obese mice. Cell 138, 961-975 (2009). 82. Mowers, J., et al. Inflammation produces catecholamine resistance in obesity via activation of PDE3B by the protein kinases IKKepsilon and TBK1. Elife 2, e01119 (2013). 83. Shi, S.Y., et al. JAK2 promotes brown adipose tissue function and is required for diet- and cold-induced thermogenesis in mice. Diabetologia 59, 187-196 (2016). 84. Hagman, J. Critical Functions of IRF4 in B and T Lymphocytes. J Immunol 199, 3715-3716 (2017). 85. Eguchi, J., et al. Transcriptional control of adipose lipid handling by IRF4. Cell Metab 13, 249-259 (2011). 86. Kissig, M., et al. PRDM16 represses the type I interferon response in adipocytes to promote mitochondrial and thermogenic programing. EMBO J 36, 1528-1542 (2017).

87. Cappelletti, M., Lawson, M.J., Chan, C.C., Wilburn, A.N. & Divanovic, S. Differential outcomes of TLR2 engagement in inflammation-induced preterm birth. J Leukoc Biol (2017). 88. Cappelletti, M., et al. Type I interferons regulate susceptibility to inflammation- induced preterm birth. JCI Insight 2, e91288 (2017). 89. Thomas, K.E., Galligan, C.L., Newman, R.D., Fish, E.N. & Vogel, S.N. Contribution of interferon-beta to the murine macrophage response to the toll-like receptor 4 agonist, lipopolysaccharide. J Biol Chem 281, 31119-31130 (2006). 90. Young, P., Arch, J.R. & Ashwell, M. Brown adipose tissue in the parametrial fat pad of the mouse. FEBS Lett 167, 10-14 (1984). 91. Loncar, D., et al. The effect of intermittent cold treatment on the adipose tissue of the cat. Apparent transformation from white to brown adipose tissue. J Ultrastruct Mol Struct Res 97, 119-129 (1986). 92. Lee, M.W., et al. Activated type 2 innate lymphoid cells regulate beige fat biogenesis. Cell 160, 74-87 (2015). 93. Qiu, Y., et al. Eosinophils and type 2 cytokine signaling in macrophages orchestrate development of functional beige fat. Cell 157, 1292-1308 (2014). 94. Fischer, K., et al. Alternatively activated macrophages do not synthesize catecholamines or contribute to adipose tissue adaptive thermogenesis. Nat Med 23, 623-630 (2017). 95. Babaei, R., et al. Jak-TGFbeta cross-talk links transient adipose tissue inflammation to beige adipogenesis. Sci Signal 11(2018). 96. Chung, K.J., et al. A self-sustained loop of inflammation-driven inhibition of beige adipogenesis in obesity. Nat Immunol 18, 654-664 (2017). 97. Haslam, D.W. & James, W.P. Obesity. Lancet 366, 1197-1209 (2005). 98. Falagas, M.E. & Kompoti, M. Obesity and infection. Lancet Infect Dis 6, 438-446 (2006). 99. Giles, D.A., et al. Thermoneutral housing exacerbates nonalcoholic fatty liver disease in mice and allows for sex-independent disease modeling. Nat Med 23, 829-838 (2017). 100. Harley, I.T., et al. IL-17 signaling accelerates the progression of nonalcoholic fatty liver disease in mice. Hepatology 59, 1830-1839 (2014). 101. Braunersreuther, V., Viviani, G.L., Mach, F. & Montecucco, F. Role of cytokines and chemokines in non-alcoholic fatty liver disease. World J Gastroenterol 18, 727- 735 (2012). 102. Mukherjee, R., et al. Nicotinamide adenine dinucleotide phosphate (reduced) oxidase 2 modulates inflammatory vigor during nonalcoholic fatty liver disease progression in mice. Hepatol Commun 2, 546-560 (2018). 103. Giles, D.A., et al. Regulation of Inflammation by IL-17A and IL-17F Modulates Non- Alcoholic Fatty Liver Disease Pathogenesis. PLoS One 11, e0149783 (2016). 104. Robert, M. & Miossec, P. Effects of Interleukin 17 on the cardiovascular system. Autoimmun Rev 16, 984-991 (2017). 105. Mehra, V.C., Ramgolam, V.S. & Bender, J.R. Cytokines and cardiovascular disease. J Leukoc Biol 78, 805-818 (2005).

106. Giles, D.A., et al. Modulation of ambient temperature promotes inflammation and initiates atherosclerosis in wild type C57BL/6 mice. Mol Metab 5, 1121-1130 (2016). 107. McCall, K.D., et al. Phenylmethimazole blocks palmitate-mediated induction of inflammatory cytokine pathways in 3T3L1 adipocytes and RAW 264.7 macrophages. J Endocrinol 207, 343-353 (2010). 108. Ghazarian, M., et al. Type I Interferon Responses Drive Intrahepatic T cells to Promote Metabolic Syndrome. Sci Immunol 2(2017). 109. Wang, X.A., et al. Interferon regulatory factor 7 deficiency prevents diet-induced obesity and insulin resistance. Am J Physiol Endocrinol Metab 305, E485-495 (2013). 110. Wang, X.A., et al. Interferon regulatory factor 3 constrains IKKbeta/NF-kappaB signaling to alleviate hepatic steatosis and insulin resistance. Hepatology 59, 870- 885 (2014). 111. Kumari, M., et al. IRF3 promotes adipose inflammation and insulin resistance and represses browning. J Clin Invest 126, 2839-2854 (2016). 112. Franceschini, L., et al. Reciprocal interference between insulin and interferon- alpha signaling in hepatic cells: a vicious circle of clinical significance? Hepatology 54, 484-494 (2011). 113. Bhattacharya, A., et al. Superoxide Dismutase 1 Protects Hepatocytes from Type I Interferon-Driven Oxidative Damage. Immunity 43, 974-986 (2015). 114. Koivisto, V.A., Pelkonen, R. & Cantell, K. Effect of interferon on glucose tolerance and insulin sensitivity. Diabetes 38, 641-647 (1989). 115. Tremlett, H.L., Yoshida, E.M. & Oger, J. Liver injury associated with the beta- interferons for MS: a comparison between the three products. Neurology 62, 628- 631 (2004). 116. Wieser, V., et al. Adipose type I interferon signalling protects against metabolic dysfunction. Gut (2016). 117. Martens, K., Bottelbergs, A. & Baes, M. Ectopic recombination in the central and peripheral nervous system by aP2/FABP4-Cre mice: implications for metabolism research. FEBS Lett 584, 1054-1058 (2010). 118. Lee, K.Y., et al. Lessons on conditional gene targeting in mouse adipose tissue. Diabetes 62, 864-874 (2013). 119. Coelho, P., Almeida, J., Prudencio, C., Fernandes, R. & Soares, R. Effect of Adipocyte Secretome in Melanoma Progression and Vasculogenic Mimicry. J Cell Biochem 117, 1697-1706 (2016). 120. Park, J., Euhus, D.M. & Scherer, P.E. Paracrine and endocrine effects of adipose tissue on cancer development and progression. Endocr Rev 32, 550-570 (2011). 121. Moreno-Fernandez, M.E., et al. Peroxisomal beta-oxidation regulates whole body metabolism, inflammatory vigor, and pathogenesis of nonalcoholic fatty liver disease. JCI Insight 3(2018). 122. Yang, D., et al. Utilization of adipocyte-derived lipids and enhanced intracellular trafficking of fatty acids contribute to breast cancer progression. Cell Commun Signal 16, 32 (2018). 123. Balaban, S., Lee, L.S., Schreuder, M. & Hoy, A.J. Obesity and cancer progression: is there a role of fatty acid metabolism? Biomed Res Int 2015, 274585 (2015).

124. Hjuler, K.F., et al. Increased global arterial and subcutaneous adipose tissue inflammation in patients with moderate-to-severe psoriasis. Br J Dermatol 176, 732-740 (2017). 125. Guzik, T.J., Skiba, D.S., Touyz, R.M. & Harrison, D.G. The role of infiltrating immune cells in dysfunctional adipose tissue. Cardiovasc Res 113, 1009-1023 (2017). 126. Couturier, J., et al. Human adipose tissue as a reservoir for memory CD4+ T cells and HIV. AIDS 29, 667-674 (2015). 127. Hazan, U., et al. Human adipose cells express CD4, CXCR4, and CCR5 [corrected] receptors: a new target cell type for the immunodeficiency virus-1? FASEB J 16, 1254-1256 (2002). 128. Beigier-Bompadre, M., et al. Mycobacterium tuberculosis infection modulates adipose tissue biology. PLoS Pathog 13, e1006676 (2017). 129. Bouwman, J.J., Visseren, F.L., Bouter, K.P. & Diepersloot, R.J. Infection-induced inflammatory response of adipocytes in vitro. Int J Obes (Lond) 32, 892-901 (2008). 130. Onuora, S. Obesity hampers effects of anti-TNF agents. Nat Rev Rheumatol 14, 320 (2018). 131. Gupta-Ganguli, M., Cox, K., Means, B., Gerling, I. & Solomon, S.S. Does therapy with anti-TNF-alpha improve glucose tolerance and control in patients with type 2 diabetes? Diabetes Care 34, e121 (2011). 132. Beger, R.D. A review of applications of metabolomics in cancer. Metabolites 3, 552-574 (2013). 133. Jutley, G.S. & Young, S.P. Metabolomics to identify biomarkers and as a predictive tool in inflammatory diseases. Best Pract Res Clin Rheumatol 29, 770-782 (2015). 134. Santoru, M.L., et al. Cross sectional evaluation of the gut-microbiome metabolome axis in an Italian cohort of IBD patients. Sci Rep 7, 9523 (2017). 135. Shommu, N.S., et al. The Use of Metabolomics and Inflammatory Mediator Profiling Provides a Novel Approach to Identifying Pediatric Appendicitis in the Emergency Department. Sci Rep 8, 4083 (2018). 136. Young, S.P., et al. The impact of inflammation on metabolomic profiles in patients with arthritis. Arthritis Rheum 65, 2015-2023 (2013). 137. Cirulli, E.T., et al. Profound Perturbation of the Metabolome in Obesity Is Associated with Health Risk. Cell Metab 29, 488-500 e482 (2019). 138. Gonzalez-Navajas, J.M., Lee, J., David, M. & Raz, E. Immunomodulatory functions of type I interferons. Nat Rev Immunol 12, 125-135 (2012). 139. Mueller, W.M., et al. Evidence that glucose metabolism regulates leptin secretion from cultured rat adipocytes. Endocrinology 139, 551-558 (1998). 140. Shen, J., Sakaida, I., Uchida, K., Terai, S. & Okita, K. Leptin enhances TNF-alpha production via p38 and JNK MAPK in LPS-stimulated Kupffer cells. Life Sci 77, 1502-1515 (2005).

Chapter 2. Obesity and its Association with Inflammation

OBESITY PANDEMIC

AT/adipocytes are appreciated regulators of overall metabolic health. Adipocytes are suggested players across various diseases including obesity1-3, autoimmunity4,5, infectious disease6-8, and cancer9,10. Naturally, our understanding of the pathogenic role of AT/adipocyte in obesity is the best established.

Obesity is the first order problem of the current generation. Body mass index (BMI) is utilized as a well-established correlative tool for body fat calculated based on weight and height. A person with a BMI greater than 30 is considered obese. Half a billion individuals have a BMI greater than 3011. However, estimations based on BMI are proposed to be conservative and may mask the true impact of the unabated obesity pandemic12,13. Male (35%) and female (40%) obesity rates are nearly equivalent in the

US, but an increase in prevalence is observed in females alone.14 In fact, obesity is a major clinical burden and leads to a significant loss of disease-free years15 and hampers professions requiring peak physical activity (e.g., military)18.

Among all nations, the United States leads the way with nearly 40% of the population considered obese19. The US alone contributes approximately $147 billion dollars16 per year to obesity-associated medical costs. Globally, costs are estimated to approach $2 trillion dollars17. Obesity was long perceived to be restricted to developed nations. However, developing nations are now similarly experiencing a sharp rise in obesity. Alarmingly, rates of obesity in some developing nations are outpacing developed nations.20

In conjunction, and contrary to previous beliefs, children are not spared as 19% aged 2-19 are obese19. In fact, it is predicted that ~60% of children will develop obesity in adulthood21. While environmental factors undoubtedly contribute to the prevalence of childhood obesity and its associated metabolic derangements, these effects are likely compounded by parental factors. Over 1/3 of reproductive age women in the United

States are obese22. Obesity impacts both paternal (e.g., epigenetic marks)23-25 and maternal (e.g., oocyte quality)26-29 reproductive determinants. Children born to obese mothers are at significant risk for development of metabolic disease30-32. Thus, vertical transmission is a likely contributor to the vicious generational cycle that propagates obesity persistence.

Obesity is the culmination of a persistent imbalance between caloric intake (e.g., food intake), calories processed into energy, and the amount of energy stored into fat or utilized for critical daily activities. Accrual of excess energy leads to the continuous expansion of adipose tissue (AT) depots and eventual pathological accumulation33,34.

While this fundamental understanding of obesity development stems back centuries, the underlying mechanisms are highly complex and remain poorly understood. Although obesity is a multifactorial disease involving diet, genetics, and environmental factors, obesity-associated inflammation is a well-established pathogenic link to metabolic and end-organ sequelae including non-alcoholic fatty liver disease (NAFLD), cardiovascular diseases, rheumatoid arthritis, multiple sclerosis, Alzheimer’s disease, and diverse cancers35,36. The pathological accumulation of excess AT drives increases in systemic levels of various inflammatory triggers (e.g., endotoxin37, free DNA38, free fatty acids39).

Further, excessive accumulation of AT drives detrimental pathogenic consequences

34 including adipocyte hypoxia, ER stress and death that contributes to obesity-associated inflammation40-42. These collective processes augment a broad activation of inflammatory cascades that in unison coordinate obesity-associated inflammation43.

A direct consequence of obesity-associated inflammation is the subsequent increase in infiltration of diverse immune cells (e.g., neutrophils, macrophages, NK cells,

DCs, CD4 and CD8 T cells, B cells) into metabolic organs, including the AT44,45. For instance, in obese AT, the diversity of cell types greatly expands and immune cells are estimated to approach up to 40% of the total tissue, with macrophages representing the predominant immune cell type5,44. Activation of these infiltrating immune cells directly modulate metabolic health (e.g., insulin sensitivity, hepatocellular death and steatohepatitis)43. Obesity likewise enhances the production of proinflammatory mediators, classically TNF46 and IL-647, that induce metabolic derangements48-50. Beyond these traditional studied proinflammatory cytokines, recent evidence indicates that a plethora of proinflammatory mediators, including type I interferons (IFN)51,52 and B cell activating factor (BAFF)53-56, contributes to obesity-associated sequelae. Thus, much effort has been placed on targeting obesity-associated inflammation to dampen disease pathogenesis. As much of our understanding is based on the contribution of immune cells to obesity-associated inflammation, a greater definition of the contribution of non- hematopoietic cells (e.g., adipocytes) are fully warranted and may yield new and appealing therapeutic targets.

OVERVIEW OF TYPE I INTERFERONS

Type I IFN are secreted peptides with multifaceted roles governing innate and adaptive immune responses57. Definition of type I IFNs in context of viral infections was the initial predominant area of study. However, it is now well appreciated that the impact of type I

IFNs is broadly implicated across human health and disease including bacterial infections/sepsis58, autoimmunity59, cancer60, and most recently metabolism51,52. Type I

IFNs exhibit both beneficial and detrimental effects. Activation of type I IFNs provides necessary defensive support against invading pathogens58. In contrast, persistent activation leads to detrimental consequences including autoimmunity59 and immune suppression58. Deeper examination of type I IFNs in these areas have collectively contributed to a greater definition of the mechanisms governing type I IFN regulation.

Over a dozen type I IFN subtypes exist in mice and humans (e.g., IFNα, IFNβ,

IFNω, IFNε, IFNτ, and IFNκ). IFNα (14 subtypes) and IFNβ (1 subtype) are the most well expressed, conserved across species (e.g., mouse and human) and characterized.61,62

Other described type I IFN subtypes appear to play different roles across species.62

Although IFNβ is widely expressed, IFNα production is largely restricted to hematopoietic cells.63 Among type I IFN producing cell types, plasmacytoid DCs (pDC) are deemed

“natural IFN producing” cells due to their ability to produce large quantities.64 Across cells types, the predominant means of type I IFN induction involve extracellular or intracellular pattern recognition receptors (PRR), such as toll-like receptors (TLRs) or RIG-I receptors, that detect pathogen associated molecular patterns (PAMPs).63 PRR sensing of PAMPs drives signaling through pathways including TIR-domain containing adapter-inducing interferon beta (TRIF)/tumor necrosis factor receptor-associated factors (TRAF; e.g.,

TLRs), MAVS (e.g., RIG-I, NOD2) or STING (e.g., cytosolic DNA sensors).58 These diverse pathways in turn activate interferon regulatory factor (IRF) 3 or 7 to induce transcription of type I IFN (Fig. 1).63 Type I IFN production is amplified by a positive feedback loop (Fig. 1), in part, by the ability of type I IFNs to induce IRF7.65 Regulation of type I IFN production and how it relates to disease pathogenesis (e.g., infection, SLE) is still intensely investigated.

Beyond PRR activation, recent findings suggest that endogenous cytokines (e.g.,

TNF66,67 and RANKL68) can likewise induce type I IFN production. Distinctly, these endogenous cytokines drive type I IFN production through IRF166,67. However, the mechanisms governing this means of type I IFN induction remain poorly understood and are under investigation.

Type I IFN effector capacity is likewise a highly regulated process. Type I IFNs bind and activate the ubiquitously expressed heterodimeric receptor type I IFN α/β receptor (IFNAR) composed of the IFNAR1 and IFNAR2 chains (Fig. 1). Described nearly three decades ago, canonical IFNAR-signaling cascade triggers Janus kinases (i.e., Tyk2 and Jak1). Tyrosine phosphorylation of Signal Transducer and Activator Transcription proteins (STAT; i.e., STAT1/STAT2) by Jak1/Tyk2 facilitates STAT1/2 multimerization and translocation to the nucleus wherein STAT1/2 further assembles with IRF9 – collectively known as IFN-stimulated gene factor 3 (ISGF3). ISGF3 recognizes IFN- stimulated response elements (ISRE) of hundreds of interferon stimulated genes (ISGs) to promote their transcription. Notably ISRE sequences (TTTCNNTTTC) are distinct from other cytokine-driven STAT homodimer binding elements (TTCNNNGAA), which allows for additional layers of regulation. ISG-encoded proteins can restrict pathogens via

37 impeding their replication.69 The mechanisms governing the switch between basal/low levels to amplified levels of type I IFN production is still explored.

Basal type I IFN expression is posited to allow for rapid and effective antimicrobial programming in immune cells.57 Notably picomolar quantities of type I IFNs are sufficient to prime macrophages for enhanced responsiveness.57 Homeostatic type I IFN production is regulated partly via gut commensals.70,71 Histone deacetylase 3 (HDCA3) is a key regulator of type I IFN expression.72 As establishment of the gut microbiome is interlinked with HDAC3 activity73, the interplay between basal type I IFN production, microbiome and HDAC3 is an intriguing question. Further, HDAC3 is a proposed brake of white adipose tissue (WAT) browning.74 If type I IFNs are involved in HDAC3 control of

WAT browning or metabolic capacity of WAT is unknown but should be studied.

Basal type I IFN production is regulated by mechanisms including downregulation of IFNAR expression57, induction of negative regulators (e.g., suppressor of cytokine signaling [SOCS])75, microRNA (miRNA)76, and pausing of RNA polymerase II on IFN pathway genes57. Notably, pathogens have evolved mechanisms to exploit these negative regulators of type I IFNs to evade host antimicrobial responses77 – whether exploitation of these mechanisms could be utilized as therapeutic approaches for pathogenic type I IFN-driven diseases is an attractive area of study.

Beyond control of basal type I IFN secretion, in necessary moments (e.g., control of pathogen burden), mechanisms exist to allow for significant boosts to type I IFN production. Type I IFN production facilitates a positive feedback loop, thereby providing one means of amplification. Enhancement of STAT1 activation, by other pro-inflammatory cytokines (e.g., IL-6, IFNγ), and IRF9 similarly facilitate boosting of the type I IFN/IFNAR

38 axis. Augmented expression of STAT1/IRF9 is capable of maintaining expression of ISGs over long periods of time78. However, a persistent boost of type I IFN production has unintended consequences, specifically in driving disease pathology. Many risk variants exist of signature genes associated with the type I IFN pathway including IRFs79, ISGs59,

STATs80, and MAVS81. It is proposed that these risk variants influence B cell activation59.

The direct mechanisms interlinking these risk factors with autoimmunity development and pathogenesis remain elusive and under investigation.

Type I IFN axis potently modulates immune cell activation and function across innate and adaptive immunity. In addition to the direct antimicrobial effects of type I IFN induced ISGs mentioned above, type I IFNs upregulate MHC class I, MHC class II, CD80,

CD86 expression to enhance antigen recognition by myeloid cells.82 Type I IFNs likewise can augment differentiation of monocytes into DCs83, and enhance DC and macrophage cytokine production potential (e.g., TNF and IL6)84,85. Type I IFNs can also influence antigen presenting cell (APCs) recruitment to target tissues via upregulation of chemokine receptors on monocytes (e.g., CCR7) or induction of chemokines (e.g., CXCL9, CXCL10,

CCL19)86,87. Control of macrophage phagocytosis and oxidative burst are similarly modulated by type I IFNs88. Further, type I IFNs can regulate NK cell capacity for direct target cell killing and IFN-γ production89. While the precise mechanisms underlying these type I IFN driven effects remain under intense study, hints have emerged suggesting alteration of core cellular metabolism90 and epigenetics91 are integral.

Type I IFNs broadly impact adaptive immune cell responses including T and B cells. Polarization of CD4 T cells can be directly skewed towards Th1-like cells by type I

IFNs92. Contrasting reports exist on type I IFN effects on Th2 cells – some suggesting

39 that type I IFNs prohibit Th2 polarization93, while others indicate that Th2 induction by

DCs require type I IFN94. Greater elucidation of the exact role of type I IFNs in CD4 T cell polarization are thus required. Type I IFNs also impact CD8 T cell effector capacity. In vitro95,96 and in vivo97,98 studies suggest that type I IFNs promote CD8 T cell survival and

IFN-γ production. Type I IFNs directly sustain expression of T-bet and Eomes to support

CD8 T cell activation and differentiation99. Further, type I IFNs prime naϊve CD8 T cells, allowing more rapid gain of effector functions upon antigen stimulation100. Direct effects of type I IFNs on B cells include enhancement of primary antibody response, memory, production of all IgG subtypes, and modulation of long-lived antibodies101. Notably, type I

IFN-driven isotype switching is promoted via its impact on DCs101. These effects on B cells are hypothesized to be critical for local antiviral humoral responses102. Combined, this evidence highlights an integral role for type I IFNs to modify and augment adaptive immune responses.

As type I IFNs are multifaceted mediators of innate and adaptive immune responses, many mechanisms exist to allow for fine control over its pleiotropic effects across immune cells. Magnitude (e.g., positive feedback loop), as previously discussed, is one means by which type I IFNs control IFNAR signaling. Additionally, the type I

IFN/IFNAR axis differentially activates STATs to activate distinct target genes and promote specific functions. While IFNAR can activate STAT1 and STAT3 in most cells,

STAT4, STAT5, and STAT6 activation is restricted to certain cell types.103 Priming effects described above for instance may rely more heavily on STAT1104. Cooperation with other

TFs, including IRFs, add additional layers of complexity to type I IFN/IFNAR axis regulation. This coordination facilitates activation of ISGs that drive unique effector

40 responses105,106. An improved comprehension of these mechanisms driving distinctive type I IFN effects in different cells types could yield highly specific, and appealing, therapeutic targets. Given that adipocytes possess some “immune-like” function (e.g.,

TLR expression, cytokine production, antigen presentation), whether the type I

IFN/IFNAR axis can modify adipocyte inflammatory potential is unknown. This foundational question forms the basis of the interrogation and observations presented in

Chapter 3 below.

Figure 1. Type I IFN/IFNAR Axis. Type I IFN production is regulated by various pathways.

Recognition of pathogen associated molecular patterns (PAMPs) by pattern recognition receptors

(PRR) including TLRs, NOD, RIG-I, and cytosolic DNA sensors (cGAS) signaling mediate interferon regulatory factor (IRF) 3/IRF7 induction of type I IFNs. Host factors, including cytokines

(e.g., TNF), can alternatively induce type I IFN production through IRF1. Type I IFNs activate the heterodimeric type I IFN α/β receptor (IFNAR). IFNAR signaling is mediated canonically through

STAT1/STAT2/IRF9 (collectively known as ISGF3) complex. ISGF3 promote the transcription of interferon-sensitive response elements (ISRE; also known as interferon stimulated genes [ISG]) resulting in pleiotropic effects driven by type I IFNs. ISGF3 can also induce type I IFN production, thereby driving a positive feedback loop.

OVERVIEW OF B CELL ACTIVATING FACTOR (BAFF) AND A PROLIFERATION-

INDUCING LIGAND (APRIL) AXIS

The TNF superfamily consists of 19 identified members and 29 associated receptors.107

Among TNF superfamily members, B-cell activating factor (BAFF; also known as BLyS,

TNSF13B, TALL1) and A proliferation-inducing ligand (APRIL; also known as TRDL-1,

TNSF13) are close homologs (33% AA similarity) that share some biological activities including maturation and survival of B cells108,109. BAFF (32 kDa) and APRIL (27.4 kDa) are molecules of similar molecular weight and produced by diverse hematopoietic (e.g., macrophages, DCs, T cells, B cells)108,109 and non-hematopoietic cells (e.g., adipocytes110, astrocytes111, gut epithelial cells112). Maintenance of B cell homeostasis depends on radio-resistant stromal cell production of BAFF, but not bone-marrow derived cells.113 Reasons for this are poorly understood. Both BAFF and APRIL exist as membrane bound114,115 and soluble forms116. Soluble BAFF and APRIL can exist as oligomers and homo/hetero multimers of BAFF/APRIL.116 This complexity may allow for fine tuning of BAFF and APRIL responses. BAFF and APRIL expression is augmented by multiple proinflammatory mechanisms including type I IFNs, IFNγ, and TLRs.117

However, regulation of BAFF and APRIL production, in health and disease, remains a critical question and under heavy investigation.

BAFF and APRIL share two receptors, transmembrane activator and CAML interactor (TACI; also known as TNFRSF13b) and B cell maturation antigen (BCMA; also known as TNFRSF17), and BAFF alone binds an additional receptor, BAFF receptor

(BAFF-R; also known as TNFRSF13C) (Fig. 2).118 Notably, discovery of BCMA and TACI preceded the characterization of BAFF and APRIL. BCMA was originally identified in a

43 human lymphoma cell line and originally its expression was believed to be relatively restricted to B cells.119 TACI was described as a calcium-modulator in T and B cells.120

After the identification of BAFF, discovery of BAFF-R was spawned by the observation that a BAFF binding cell line (e.g., BJAB) lacked BCMA and TACI expression.121 Although all 3 receptors are primarily expressed on B cells, expression of these receptors has been also observed in non-hematopoietic cells (e.g., adipocytes110). TACI and BCMA signaling is mediated by canonical NF-κB (e.g., TRAF1/2). BAFF-R signaling is driven through non- canonical NF-κB cascade (e.g., TRAF3/6). Conformation of soluble BAFF or APRIL oligomers or multimers dictates the selectivity for each receptor. For instance, trimers of

BAFF can bind to all 3 receptors, whereas APRIL-dominant multimer (e.g.,

BAFF/APRIL/APRIL) binds to TACI and BCMA only.116 It is suggested that while various conformations of BAFF and APRIL can bind to its cognate receptor, receptor activation requires specific combinations.116,122 The physiological consequences of this level of regulation remains poorly understood and may reveal more targeted approaches of the

BAFF/APRIL system.

The role of BAFF, APRIL, and its 3 receptors have been elucidated primarily through gain/loss of function models in mice. Genetic deletion of BAFF and BAFF-R result in drastic reduction in mature B cell pool.121,123 In contrast, BAFF-Tg mice exhibit expansion of the B cell compartment and increased production of antinuclear autoantibodies.124 Genetic loss of APRIL, TACI, and BCMA on mature B cell numbers is not as pronounced as seen with BAFF or BAFF-R.118 Despite apparent lack of impact on

B cell numbers, APRIL, TACI, and BCMA modulate various aspects of B cell function.

APRIL and TACI mediate isotype switching in B cells.125,126 BCMA upregulates B cell

44 antigen presentation capacity.127 Further BCMA is a critical mediator of plasma cell survival.128 Despite most studies examining the function of BAFF/APRIL receptors independently via single receptor deletion, interplay likely exists among all 3 receptors.

Consequences of parallel receptor activation and mechanisms regulating their interrelated function(s) should be further explored.

As manipulation of the BAFF/APRIL system in mice revealed critical roles in B cell maturation, survival, and function, it is thus unsurprising that BAFF/APRIL activity is tied to disease. BAFF/APRIL have implicated roles in various diseases, including cancer (e.g., myeloma) and infectious diseases, however the role of BAFF/APRIL in autoimmunity remains the most well defined. BAFF-Tg mice exhibit severe autoimmunity, with characteristics similar to SLE and Sjögren’s syndrome.129 Blockade of BAFF in SLE mouse models dampens disease.130 These bench observations have yielded valuable bedside approaches. BAFF and APRIL levels are significantly higher in patients with autoimmune disease (e.g., SLE, Sjögren’s, RA). Despite this correlation, the recommendations on the use of BAFF and APRIL as predictive markers for disease severity remains divided. Some have demonstrated that elevated BAFF levels are prognostic markers for moderate/severe SLE131 and predictive of SLE flares132. In contrast, other reports indicate no association between serum BAFF and autoimmune severity133 nor prediction for responsiveness to anti-BAFF therapy134. The exact reasons for these diverse findings remain elusive. However, more nuanced examination of BAFF and APRIL levels to markers of disease (e.g., renal activity, CNS disease) is clearly needed.

Belimumab, a human recombinant monoclonal antibody against soluble BAFF, in combination with standard therapy (e.g., steroids) has undergone phase 3 trials and seen the most success in SLE treatment.135 Belimumab reduced circulating naïve B cells, activated B cells and plasma cells, but not memory B cells, and lowered autoantibody levels.136 Notably, patients with the highest disease activity and most significant levels of autoantibodies respond best to Belimumab137. As understanding of the BAFF/APRIL system has greatly expanded, new therapies have emerged from the pipeline including

Tabalumab and Blisibimod (neutralize both soluble and membrane bound BAFF),

Atacicept (antagonist of TACI and BCMA)138, and BAFF-R inhibiting drugs139. Innovative strategies including CAR T-cell targeting BAFF-R are in development140. Although therapeutics targeting the BAFF/APRIL system are exclusively aimed at reducing bioactivity of this axis, the BAFF and APRIL axis likewise play vital roles in maintenance of human health (e.g., antimicrobial, tolerance) and their therapeutic potential should be explored. Likewise, expression of BAFF, APRIL and its 3 receptors in non-hematopoietic cells, including adipocytes, present a lot of exciting questions and possibilities to be pursued. Whether the BAFF/APRIL axis modifies adipocyte homeostasis has not been previously explored and this overarching question forms the basis of studies in Chapter

Figure 2. BAFF/APRIL axis. B cell activating factor (BAFF) and A proliferation-inducing ligand

(APRIL) are close homologs within the TNF superfamily with some overlapping function (e.g., B cell biology). Both BAFF and APRIL can bind to transmembrane Activator and CAML interactor

(TACI) and B-cell maturation antigen (BCMA), while BAFF can also binds to BAFF receptor

(BAFF-R). BAFF and APRIL exist as membrane bound and soluble molecules. Soluble BAFF and

APRIL can exist as oligomers or homo/hetero multimers. Conformation of BAFF/APRIL multimers regulates the affinity to which they can bind to a specific receptor. Typically, BAFF dominant conformations (e.g., BAFF-mer, BAFF/BAFF/APRIL) can activate all three receptors. In contrast,

APRIL dominant conformations (e.g., APRIL-mer, BAFF/APRIL/APRIL) have high affinity for TACI and BCMA, but very weak binding capacity to BAFF-R. TACI and BCMA signaling is mediated by a TNF receptor associated factors (TRAF) 1/TRAF2 dependent canonical NF-κB pathway. On the otherhand, BAFF-R signaling drives a TRAF3/TRAF6 mediated non-canonical NF-κB cascade.

REFERENCES

1. Tandon, P., Wafer, R. & Minchin, J.E.N. Adipose morphology and metabolic disease. J Exp Biol 221(2018). 2. Grant, R.W. & Dixit, V.D. Adipose tissue as an immunological organ. Obesity (Silver Spring) 23, 512-518 (2015). 3. Ghigliotti, G., et al. Adipose tissue immune response: novel triggers and consequences for chronic inflammatory conditions. Inflammation 37, 1337-1353 (2014). 4. Hjuler, K.F., et al. Increased global arterial and subcutaneous adipose tissue inflammation in patients with moderate-to-severe psoriasis. Br J Dermatol 176, 732-740 (2017). 5. Guzik, T.J., Skiba, D.S., Touyz, R.M. & Harrison, D.G. The role of infiltrating immune cells in dysfunctional adipose tissue. Cardiovasc Res 113, 1009-1023 (2017). 6. Desruisseaux, M.S., Nagajyothi, Trujillo, M.E., Tanowitz, H.B. & Scherer, P.E. Adipocyte, adipose tissue, and infectious disease. Infect Immun 75, 1066-1078 (2007). 7. Beigier-Bompadre, M., et al. Mycobacterium tuberculosis infection modulates adipose tissue biology. PLoS Pathog 13, e1006676 (2017). 8. Bouwman, J.J., Visseren, F.L., Bouter, K.P. & Diepersloot, R.J. Infection-induced inflammatory response of adipocytes in vitro. Int J Obes (Lond) 32, 892-901 (2008). 9. Coelho, P., Almeida, J., Prudencio, C., Fernandes, R. & Soares, R. Effect of Adipocyte Secretome in Melanoma Progression and Vasculogenic Mimicry. J Cell Biochem 117, 1697-1706 (2016). 10. Park, J., Euhus, D.M. & Scherer, P.E. Paracrine and endocrine effects of adipose tissue on cancer development and progression. Endocr Rev 32, 550-570 (2011). 11. Bhaskaran, K., Dos-Santos-Silva, I., Leon, D.A., Douglas, I.J. & Smeeth, L. Association of BMI with overall and cause-specific mortality: a population-based cohort study of 3.6 million adults in the UK. Lancet Diabetes Endocrinol 6, 944-953 (2018). 12. Blijdorp, K., et al. Obesity is underestimated using body mass index and waist-hip ratio in long-term adult survivors of childhood cancer. PLoS One 7, e43269 (2012). 13. Robinson, E. Overweight but unseen: a review of the underestimation of weight status and a visual normalization theory. Obes Rev 18, 1200-1209 (2017). 14. Flegal, K.M., Kruszon-Moran, D., Carroll, M.D., Fryar, C.D. & Ogden, C.L. Trends in Obesity Among Adults in the United States, 2005 to 2014. JAMA 315, 2284- 2291 (2016). 15. Nyberg, S.T., et al. Obesity and loss of disease-free years owing to major non- communicable diseases: a multicohort study. Lancet Public Health 3, e490-e497 (2018). 16. Finkelstein, E.A., Trogdon, J.G., Cohen, J.W. & Dietz, W. Annual medical spending attributable to obesity: payer-and service-specific estimates. Health Aff (Millwood) 28, w822-831 (2009).

17. Tremmel, M., Gerdtham, U.G., Nilsson, P.M. & Saha, S. Economic Burden of Obesity: A Systematic Literature Review. Int J Environ Res Public Health 14(2017). 18. Breland, J.Y., et al. The Obesity Epidemic in the Veterans Health Administration: Prevalence Among Key Populations of Women and Men Veterans. J Gen Intern Med 32, 11-17 (2017). 19. Hales, C.M., Carroll, M.D., Fryar, C.D. & Ogden, C.L. Prevalence of Obesity Among Adults and Youth: United States, 2015-2016. NCHS Data Brief, 1-8 (2017). 20. Bhurosy, T. & Jeewon, R. Overweight and obesity epidemic in developing countries: a problem with diet, physical activity, or socioeconomic status? ScientificWorldJournal 2014, 964236 (2014). 21. Biro, F.M. & Wien, M. Childhood obesity and adult morbidities. Am J Clin Nutr 91, 1499S-1505S (2010). 22. Collaboration, N.R.F. Trends in adult body-mass index in 200 countries from 1975 to 2014: a pooled analysis of 1698 populations-based measurement studies with 19.2 million participants. Lancet 387, 1377-1396 (2016). 23. Soubry, A., et al. Paternal obesity is associated with IGF2 hypomethylation in newborns: results from a Newborn Epigenetics Study (NEST) cohort. BMC Med 11, 29 (2013). 24. Ost, A., et al. Paternal diet defines offspring chromatin state and intergenerational obesity. Cell 159, 1352-1364 (2014). 25. Fullston, T., et al. Paternal obesity induces metabolic and sperm disturbances in male offspring that are exacerbated by their exposure to an "obesogenic" diet. Physiol Rep 3(2015). 26. Jungheim, E.S., et al. Diet-induced obesity model: abnormal oocytes and persistent growth abnormalities in the offspring. Endocrinology 151, 4039-4046 (2010). 27. Ge, Z.J., et al. DNA methylation in oocytes and liver of female mice and their offspring: effects of high-fat-diet-induced obesity. Environ Health Perspect 122, 159-164 (2014). 28. Igosheva, N., et al. Maternal diet-induced obesity alters mitochondrial activity and redox status in mouse oocytes and zygotes. PLoS One 5, e10074 (2010). 29. Wu, L.L., et al. Mitochondrial dysfunction in oocytes of obese mothers: transmission to offspring and reversal by pharmacological endoplasmic reticulum stress inhibitors. Development 142, 681-691 (2015). 30. Catalano, P.M. & Ehrenberg, H.M. The short- and long-term implications of maternal obesity on the mother and her offspring. BJOG 113, 1126-1133 (2006). 31. O'Reilly, J.R. & Reynolds, R.M. The risk of maternal obesity to the long-term health of the offspring. Clin Endocrinol (Oxf) 78, 9-16 (2013). 32. Gaillard, R. Maternal obesity during pregnancy and cardiovascular development and disease in the offspring. Eur J Epidemiol 30, 1141-1152 (2015). 33. Lee, M.J., Wu, Y. & Fried, S.K. Adipose tissue remodeling in pathophysiology of obesity. Curr Opin Clin Nutr Metab Care 13, 371-376 (2010). 34. Choe, S.S., Huh, J.Y., Hwang, I.J., Kim, J.I. & Kim, J.B. Adipose Tissue Remodeling: Its Role in Energy Metabolism and Metabolic Disorders. Front Endocrinol (Lausanne) 7, 30 (2016). 35. Haslam, D.W. & James, W.P. Obesity. Lancet 366, 1197-1209 (2005).

36. Strissel, K.J., et al. Adipocyte death, adipose tissue remodeling, and obesity complications. Diabetes 56, 2910-2918 (2007). 37. Cani, P.D., et al. Metabolic endotoxemia initiates obesity and insulin resistance. Diabetes 56, 1761-1772 (2007). 38. Nishimoto, S., et al. Obesity-induced DNA released from adipocytes stimulates chronic adipose tissue inflammation and insulin resistance. Sci Adv 2, e1501332 (2016). 39. Boden, G. Obesity and free fatty acids. Endocrinol Metab Clin North Am 37, 635- 646, viii-ix (2008). 40. Kawasaki, N., Asada, R., Saito, A., Kanemoto, S. & Imaizumi, K. Obesity-induced endoplasmic reticulum stress causes chronic inflammation in adipose tissue. Sci Rep 2, 799 (2012). 41. Lawler, H.M., et al. Adipose Tissue Hypoxia, Inflammation, and Fibrosis in Obese Insulin-Sensitive and Obese Insulin-Resistant Subjects. J Clin Endocrinol Metab 101, 1422-1428 (2016). 42. Ryan, S. Adipose tissue inflammation by intermittent hypoxia: mechanistic link between obstructive sleep apnoea and metabolic dysfunction. J Physiol 595, 2423- 2430 (2017). 43. Saltiel, A.R. & Olefsky, J.M. Inflammatory mechanisms linking obesity and metabolic disease. J Clin Invest 127, 1-4 (2017). 44. Weisberg, S.P., et al. Obesity is associated with macrophage accumulation in adipose tissue. J Clin Invest 112, 1796-1808 (2003). 45. Ferrante, A.W., Jr. The immune cells in adipose tissue. Diabetes Obes Metab 15 Suppl 3, 34-38 (2013). 46. Hotamisligil, G.S., Murray, D.L., Choy, L.N. & Spiegelman, B.M. Tumor necrosis factor alpha inhibits signaling from the insulin receptor. Proc Natl Acad Sci U S A 91, 4854-4858 (1994). 47. de Luca, C. & Olefsky, J.M. Inflammation and insulin resistance. FEBS Lett 582, 97-105 (2008). 48. Ventre, J., et al. Targeted disruption of the tumor necrosis factor-alpha gene: metabolic consequences in obese and nonobese mice. Diabetes 46, 1526-1531 (1997). 49. Ouchi, N., Parker, J.L., Lugus, J.J. & Walsh, K. Adipokines in inflammation and metabolic disease. Nat Rev Immunol 11, 85-97 (2011). 50. Mas, E., et al. IL-6 deficiency attenuates murine diet-induced non-alcoholic steatohepatitis. PLoS One 4, e7929 (2009). 51. Ghazarian, M., et al. Type I Interferon Responses Drive Intrahepatic T cells to Promote Metabolic Syndrome. Sci Immunol 2(2017). 52. Wieser, V., et al. Adipose type I interferon signalling protects against metabolic dysfunction. Gut 67, 157-165 (2018). 53. Kim, D.H. & Do, M.S. BAFF knockout improves systemic inflammation via regulating adipose tissue distribution in high-fat diet-induced obesity. Exp Mol Med 47, e129 (2015). 54. Tsiantoulas, D., et al. BAFF Neutralization Aggravates Atherosclerosis. Circulation (2018).

55. Kawasaki, K., et al. Blockade of B-cell-activating factor signaling enhances hepatic steatosis induced by a high-fat diet and improves insulin sensitivity. Lab Invest 93, 311-321 (2013). 56. Liu, L., et al. TACI-Deficient Macrophages Protect Mice Against Metaflammation and Obesity-Induced Dysregulation of Glucose Homeostasis. Diabetes 67, 1589- 1603 (2018). 57. Ivashkiv, L.B. & Donlin, L.T. Regulation of type I interferon responses. Nat Rev Immunol 14, 36-49 (2014). 58. McNab, F., Mayer-Barber, K., Sher, A., Wack, A. & O'Garra, A. Type I interferons in infectious disease. Nat Rev Immunol 15, 87-103 (2015). 59. Muskardin, T.L.W. & Niewold, T.B. Type I interferon in rheumatic diseases. Nat Rev Rheumatol 14, 214-228 (2018). 60. Parker, B.S., Rautela, J. & Hertzog, P.J. Antitumour actions of interferons: implications for cancer therapy. Nat Rev Cancer 16, 131-144 (2016). 61. Decker, T., Muller, M. & Stockinger, S. The yin and yang of type I interferon activity in bacterial infection. Nat Rev Immunol 5, 675-687 (2005). 62. Pestka, S., Krause, C.D. & Walter, M.R. Interferons, interferon-like cytokines, and their receptors. Immunol Rev 202, 8-32 (2004). 63. Trinchieri, G. Type I interferon: friend or foe? J Exp Med 207, 2053-2063 (2010). 64. Asselin-Paturel, C. & Trinchieri, G. Production of type I interferons: plasmacytoid dendritic cells and beyond. J Exp Med 202, 461-465 (2005). 65. Marie, I., Durbin, J.E. & Levy, D.E. Differential viral induction of distinct interferon- alpha genes by positive feedback through interferon regulatory factor-7. EMBO J 17, 6660-6669 (1998). 66. Yarilina, A., Park-Min, K.H., Antoniv, T., Hu, X. & Ivashkiv, L.B. TNF activates an IRF1-dependent autocrine loop leading to sustained expression of chemokines and STAT1-dependent type I interferon-response genes. Nat Immunol 9, 378-387 (2008). 67. Venkatesh, D., et al. Endothelial TNF receptor 2 induces IRF1 transcription factor- dependent interferon-beta autocrine signaling to promote monocyte recruitment. Immunity 38, 1025-1037 (2013). 68. Takayanagi, H., Kim, S. & Taniguchi, T. Signaling crosstalk between RANKL and interferons in osteoclast differentiation. Arthritis Res 4 Suppl 3, S227-232 (2002). 69. MacMicking, J.D. Interferon-inducible effector mechanisms in cell-autonomous immunity. Nat Rev Immunol 12, 367-382 (2012). 70. Abt, M.C., et al. Commensal bacteria calibrate the activation threshold of innate antiviral immunity. Immunity 37, 158-170 (2012). 71. Kawashima, T., et al. Double-stranded RNA of intestinal commensal but not pathogenic bacteria triggers production of protective interferon-beta. Immunity 38, 1187-1197 (2013). 72. Chen, X., et al. Requirement for the histone deacetylase Hdac3 for the inflammatory gene expression program in macrophages. Proc Natl Acad Sci U S A 109, E2865-2874 (2012). 73. Alenghat, T., et al. Histone deacetylase 3 coordinates commensal-bacteria- dependent intestinal homeostasis. Nature 504, 153-157 (2013).

74. Ferrari, A., et al. HDAC3 is a molecular brake of the metabolic switch supporting white adipose tissue browning. Nat Commun 8, 93 (2017). 75. Piganis, R.A., et al. Suppressor of cytokine signaling (SOCS) 1 inhibits type I interferon (IFN) signaling via the interferon alpha receptor (IFNAR1)-associated tyrosine kinase Tyk2. J Biol Chem 286, 33811-33818 (2011). 76. Sedger, L.M. microRNA control of interferons and interferon induced anti-viral activity. Mol Immunol 56, 781-793 (2013). 77. Liu, J., et al. Virus-induced unfolded protein response attenuates antiviral defenses via phosphorylation-dependent degradation of the type I interferon receptor. Cell Host Microbe 5, 72-83 (2009). 78. Cheon, H., et al. IFNbeta-dependent increases in STAT1, STAT2, and IRF9 mediate resistance to viruses and DNA damage. EMBO J 32, 2751-2763 (2013). 79. Niewold, T.B., et al. Association of the IRF5 risk haplotype with high serum interferon-alpha activity in systemic lupus erythematosus patients. Arthritis Rheum 58, 2481-2487 (2008). 80. Kariuki, S.N., et al. Cutting edge: autoimmune disease risk variant of STAT4 confers increased sensitivity to IFN-alpha in lupus patients in vivo. J Immunol 182, 34-38 (2009). 81. Pothlichet, J., et al. A loss-of-function variant of the antiviral molecule MAVS is associated with a subset of systemic lupus patients. EMBO Mol Med 3, 142-152 (2011). 82. Hervas-Stubbs, S., et al. Direct effects of type I interferons on cells of the immune system. Clin Cancer Res 17, 2619-2627 (2011). 83. Bogdan, C., Mattner, J. & Schleicher, U. The role of type I interferons in non-viral infections. Immunol Rev 202, 33-48 (2004). 84. Cappelletti, M., Lawson, M.J., Chan, C.C., Wilburn, A.N. & Divanovic, S. Differential outcomes of TLR2 engagement in inflammation-induced preterm birth. J Leukoc Biol (2017). 85. Cappelletti, M., et al. Type I interferons regulate susceptibility to inflammation- induced preterm birth. JCI Insight 2, e91288 (2017). 86. Parlato, S., et al. Expression of CCR-7, MIP-3beta, and Th-1 chemokines in type I IFN-induced monocyte-derived dendritic cells: importance for the rapid acquisition of potent migratory and functional activities. Blood 98, 3022-3029 (2001). 87. Antonelli, A., et al. Interferon-alpha, -beta and -gamma induce CXCL9 and CXCL10 secretion by human thyrocytes: modulation by peroxisome proliferator- activated receptor-gamma agonists. Cytokine 50, 260-267 (2010). 88. Sampson, L.L., Heuser, J. & Brown, E.J. Cytokine regulation of complement receptor-mediated ingestion by mouse peritoneal macrophages. M-CSF and IL-4 activate phagocytosis by a common mechanism requiring autostimulation by IFN- beta. J Immunol 146, 1005-1013 (1991). 89. Nguyen, K.B., et al. Coordinated and distinct roles for IFN-alpha beta, IL-12, and IL-15 regulation of NK cell responses to viral infection. J Immunol 169, 4279-4287 (2002). 90. Wu, D., et al. Type 1 Interferons Induce Changes in Core Metabolism that Are Critical for Immune Function. Immunity 44, 1325-1336 (2016).

91. Park, S.H., et al. Type I interferons and the cytokine TNF cooperatively reprogram the macrophage epigenome to promote inflammatory activation. Nat Immunol 18, 1104-1116 (2017). 92. Brinkmann, V., Geiger, T., Alkan, S. & Heusser, C.H. Interferon alpha increases the frequency of interferon gamma-producing human CD4+ T cells. J Exp Med 178, 1655-1663 (1993). 93. Dai, J., Megjugorac, N.J., Gallagher, G.E., Yu, R.Y. & Gallagher, G. IFN-lambda1 (IL-29) inhibits GATA3 expression and suppresses Th2 responses in human naive and memory T cells. Blood 113, 5829-5838 (2009). 94. Webb, L.M., et al. Type I interferon is required for T helper (Th) 2 induction by dendritic cells. EMBO J 36, 2404-2418 (2017). 95. Marrack, P., Kappler, J. & Mitchell, T. Type I interferons keep activated T cells alive. J Exp Med 189, 521-530 (1999). 96. Nguyen, K.B., et al. Critical role for STAT4 activation by type 1 interferons in the interferon-gamma response to viral infection. Science 297, 2063-2066 (2002). 97. Aichele, P., et al. CD8 T cells specific for lymphocytic choriomeningitis virus require type I IFN receptor for clonal expansion. J Immunol 176, 4525-4529 (2006). 98. Le Bon, A., et al. Direct stimulation of T cells by type I IFN enhances the CD8+ T cell response during cross-priming. J Immunol 176, 4682-4689 (2006). 99. Agarwal, P., et al. Gene regulation and chromatin remodeling by IL-12 and type I IFN in programming for CD8 T cell effector function and memory. J Immunol 183, 1695-1704 (2009). 100. Marshall, H.D., Prince, A.L., Berg, L.J. & Welsh, R.M. IFN-alpha beta and self- MHC divert CD8 T cells into a distinct differentiation pathway characterized by rapid acquisition of effector functions. J Immunol 185, 1419-1428 (2010). 101. Le Bon, A., et al. Type i interferons potently enhance humoral immunity and can promote isotype switching by stimulating dendritic cells in vivo. Immunity 14, 461- 470 (2001). 102. Coro, E.S., Chang, W.L. & Baumgarth, N. Type I IFN receptor signals directly stimulate local B cells early following influenza virus infection. J Immunol 176, 4343-4351 (2006). 103. van Boxel-Dezaire, A.H., Rani, M.R. & Stark, G.R. Complex modulation of cell type-specific signaling in response to type I interferons. Immunity 25, 361-372 (2006). 104. Tassiulas, I., et al. Amplification of IFN-alpha-induced STAT1 activation and inflammatory function by Syk and ITAM-containing adaptors. Nat Immunol 5, 1181- 1189 (2004). 105. Levy, D.E., Marie, I., Smith, E. & Prakash, A. Enhancement and diversification of IFN induction by IRF-7-mediated positive feedback. J Interferon Cytokine Res 22, 87-93 (2002). 106. Tamura, T., Yanai, H., Savitsky, D. & Taniguchi, T. The IRF family transcription factors in immunity and oncogenesis. Annu Rev Immunol 26, 535-584 (2008). 107. Aggarwal, B.B., Gupta, S.C. & Kim, J.H. Historical perspectives on tumor necrosis factor and its superfamily: 25 years later, a golden journey. Blood 119, 651-665 (2012).

108. Schneider, P., et al. BAFF, a novel ligand of the tumor necrosis factor family, stimulates B cell growth. J Exp Med 189, 1747-1756 (1999). 109. Yu, G., et al. APRIL and TALL-I and receptors BCMA and TACI: system for regulating humoral immunity. Nat Immunol 1, 252-256 (2000). 110. Alexaki, V.I., et al. Adipocytes as immune cells: differential expression of TWEAK, BAFF, and APRIL and their receptors (Fn14, BAFF-R, TACI, and BCMA) at different stages of normal and pathological adipose tissue development. J Immunol 183, 5948-5956 (2009). 111. Krumbholz, M., et al. BAFF is produced by astrocytes and up-regulated in multiple sclerosis lesions and primary central nervous system lymphoma. J Exp Med 201, 195-200 (2005). 112. Wang, Y., et al. An LGG-derived protein promotes IgA production through upregulation of APRIL expression in intestinal epithelial cells. Mucosal Immunol 10, 373-384 (2017). 113. Gorelik, L., et al. Normal B cell homeostasis requires B cell activation factor production by radiation-resistant cells. J Exp Med 198, 937-945 (2003). 114. Kowalczyk-Quintas, C., et al. Inhibition of Membrane-Bound BAFF by the Anti- BAFF Antibody Belimumab. Front Immunol 9, 2698 (2018). 115. Lee, S.M., Jeon, S.T., Suk, K. & Lee, W.H. Macrophages express membrane bound form of APRIL that can generate immunomodulatory signals. Immunology 131, 350-356 (2010). 116. Bossen, C., et al. TACI, unlike BAFF-R, is solely activated by oligomeric BAFF and APRIL to support survival of activated B cells and plasmablasts. Blood 111, 1004- 1012 (2008). 117. Mackay, F. & Schneider, P. Cracking the BAFF code. Nat Rev Immunol 9, 491- 502 (2009). 118. Mackay, F., Schneider, P., Rennert, P. & Browning, J. BAFF AND APRIL: a tutorial on B cell survival. Annu Rev Immunol 21, 231-264 (2003). 119. Laabi, Y., et al. A new gene, BCM, on chromosome 16 is fused to the interleukin 2 gene by a t(4;16)(q26;p13) translocation in a malignant T cell lymphoma. EMBO J 11, 3897-3904 (1992). 120. von Bulow, G.U. & Bram, R.J. NF-AT activation induced by a CAML-interacting member of the tumor necrosis factor receptor superfamily. Science 278, 138-141 (1997). 121. Thompson, J.S., et al. BAFF-R, a newly identified TNF receptor that specifically interacts with BAFF. Science 293, 2108-2111 (2001). 122. Chang, S.K., Mihalcik, S.A. & Jelinek, D.F. B lymphocyte stimulator regulates adaptive immune responses by directly promoting dendritic cell maturation. J Immunol 180, 7394-7403 (2008). 123. Sasaki, Y., Casola, S., Kutok, J.L., Rajewsky, K. & Schmidt-Supprian, M. TNF family member B cell-activating factor (BAFF) receptor-dependent and - independent roles for BAFF in B cell physiology. J Immunol 173, 2245-2252 (2004). 124. Mackay, F., et al. Mice transgenic for BAFF develop lymphocytic disorders along with autoimmune manifestations. J Exp Med 190, 1697-1710 (1999).

125. Castigli, E., et al. TACI and BAFF-R mediate isotype switching in B cells. J Exp Med 201, 35-39 (2005). 126. Hardenberg, G., van Bostelen, L., Hahne, M. & Medema, J.P. Thymus- independent class switch recombination is affected by APRIL. Immunol Cell Biol 86, 530-534 (2008). 127. Yang, M., et al. B cell maturation antigen, the receptor for a proliferation-inducing ligand and B cell-activating factor of the TNF family, induces antigen presentation in B cells. J Immunol 175, 2814-2824 (2005). 128. O'Connor, B.P., et al. BCMA is essential for the survival of long-lived bone marrow plasma cells. J Exp Med 199, 91-98 (2004). 129. Vincent, F.B., Morand, E.F., Schneider, P. & Mackay, F. The BAFF/APRIL system in SLE pathogenesis. Nat Rev Rheumatol 10, 365-373 (2014). 130. Gross, J.A., et al. TACI and BCMA are receptors for a TNF homologue implicated in B-cell autoimmune disease. Nature 404, 995-999 (2000). 131. Petri, M.A., et al. Baseline predictors of systemic lupus erythematosus flares: data from the combined placebo groups in the phase III belimumab trials. Arthritis Rheum 65, 2143-2153 (2013). 132. Carter, L.M., Isenberg, D.A. & Ehrenstein, M.R. Elevated serum BAFF levels are associated with rising anti-double-stranded DNA antibody levels and disease flare following B cell depletion therapy in systemic lupus erythematosus. Arthritis Rheum 65, 2672-2679 (2013). 133. Vincent, F.B., Northcott, M., Hoi, A., Mackay, F. & Morand, E.F. Association of serum B cell activating factor from the tumour necrosis factor family (BAFF) and a proliferation-inducing ligand (APRIL) with central nervous system and renal disease in systemic lupus erythematosus. Lupus 22, 873-884 (2013). 134. Wallace, D.J., et al. Safety profile of belimumab: pooled data from placebo- controlled phase 2 and 3 studies in patients with systemic lupus erythematosus. Lupus 22, 144-154 (2013). 135. Navarra, S.V., et al. Efficacy and safety of belimumab in patients with active systemic lupus erythematosus: a randomised, placebo-controlled, phase 3 trial. Lancet 377, 721-731 (2011). 136. Stohl, W., et al. Belimumab reduces autoantibodies, normalizes low complement levels, and reduces select B cell populations in patients with systemic lupus erythematosus. Arthritis Rheum 64, 2328-2337 (2012). 137. van Vollenhoven, R.F., et al. Belimumab in the treatment of systemic lupus erythematosus: high disease activity predictors of response. Ann Rheum Dis 71, 1343-1349 (2012). 138. Merrill, J.T., et al. Efficacy and Safety of Atacicept in Patients With Systemic Lupus Erythematosus: Results of a Twenty-Four-Week, Multicenter, Randomized, Double-Blind, Placebo-Controlled, Parallel-Arm, Phase IIb Study. Arthritis Rheumatol 70, 266-276 (2018). 139. McWilliams, E.M., et al. Anti-BAFF-R antibody VAY-736 demonstrates promising preclinical activity in CLL and enhances effectiveness of ibrutinib. Blood Adv 3, 447-460 (2019).

140. Qin, H., Dong, Z., Wang, X., Cheng, W., Smith, D. L., Song, J. Y., Aldoss, I., Muschen, M., Forman, S. J., & Kwak, L. W. CAR-T Cells Targeting BAFF-Receptor for B-Cell Malignancies: A Potential Alternative to CD19. Blood 130, 3180 (2017).

Chapter 3. Type I Interferon Sensing Unlocks Dormant Adipocyte Inflammatory Potential

Calvin C. Chan1,2,3, Monica Cappelletti2,3,#, Maria E. Moreno-Fernandez2,3, Traci E. Stankiewicz2,3, Michelle S.M.A. Damen2,3, Pablo C. Alarcon1,2,3, Rajib Mukherjee2,3, Rebekah Karns2,4, Matthew Weirauch2,5,6,8, Michael A. Helmrath7,9, Thomas H. Inge10 and Senad Divanovic1,2,3,11*

1Medical Scientist Training Program, Immunology Graduate Program Cincinnati Children’s Hospital Medical Center and the University of Cincinnati College of Medicine, Cincinnati, OH 45220, USA. 2Department of Pediatrics, University of Cincinnati College of Medicine, Cincinnati, OH 45220. Divisions of 3Immunobiology, 4Gastroenterology, Hepatology and Nutrition, 5Bioinformedical Informatics, 6Developmental Biology and 7Pediatric General and Thoracic Surgery, Cincinnati Children’s Hospital Medical Center, Cincinnati, OH 45229. The Centers for 8Autoimmune Genomics and Etiology, and 9Stem Cell & Organoid Medicine, Cincinnati Children’s Hospital Medical Center, Cincinnati, Ohio, USA 45229. 10Department of Surgery, Children’s Hospital Colorado, Aurora CO 80045, USA. 11Center for Inflammation and Tolerance, Cincinnati Children’s Hospital Medical Center, Cincinnati, Ohio, USA 45229.

#Current Address: Divisions of Neonatology and Developmental Biology, David Geffen School of Medicine at UCLA, Mattel Children’s Hospital UCLA, CA, USA.

INTRODUCTION

White adipose tissue (WAT) inflammation, in part via myeloid cell contribution, is central to obesity pathogenesis1,2. Despite some conserved inflammatory capabilities between myeloid cells and adipocytes3-7, mechanisms regulating adipocyte inflammatory potential remain poorly defined. Here, we show that activation of the type I interferon

(IFN)/IFNa receptor (IFNAR)8 axis amplified adipocyte inflammatory vigor and uncovered dormant gene expression patterns resembling inflammatory myeloid cells. IFNb-sensing promoted adipocyte glycolysis, while inhibition of glycolysis impeded IFNb-driven amplification of intra-adipocyte inflammation. Obesity-associated induction of the type I

IFN axis and activation of IFNAR signaling on non-hematopoietic cells strongly contributed to obesity pathogenesis in mice. Notably, IFNb effects were conserved in primary human adipocytes and detection of the type I IFN/IFNAR axis-associated signatures positively correlated with obesity-driven metabolic derangements in humans.

Collectively, our novel findings reveal a capacity for the type I IFN/IFNAR axis to uncover and regulate unifying inflammatory features in both myeloid cells and adipocytes thus hinting at a potentially underappreciated role of intra-adipocyte inflammation in disease pathogenesis.

RESULTS AND DISCUSSION

Recent findings indicating that the IFNAR axis detrimentally9-20 or beneficially21 modulates obesity-associated metabolic disease are discrepant. However, to our knowledge, the contribution of the type I IFN/IFNAR axis in regulation of WAT inflammation and adipocyte inflammatory vigor has not been examined. Given the relevance of type I IFN/IFNAR axis in regulation of myeloid cell inflammatory potential, and partial commonality in inflammatory functions between myeloid cells and adipocytes, we hypothesized that the activation of the type I IFN/IFNAR axis in adipocytes would uncover immune-like inflammatory signatures and exacerbate adipocyte intrinsic inflammatory vigor.

Obesity is associated with metabolic endotoxemia (lipopolysaccharide; LPS)22 and

LPS sensing induces type I IFN production in myeloid cells23. Type I IFN engagement of the ubiquitously expressed IFNAR initiates activation of inflammatory mediators including

STAT1, STAT3, interleukin-6 (IL-6), tumor necrosis factor (TNF) and various chemokines24. Whether LPS stimulation is sufficient to activate the type I IFN axis in primary adipocytes, as it does in myeloid cells23, has not been examined. Notably, LPS treatment of mouse primary adipocytes induced IFNb production and mRNA expression of type I IFN signature genes including Irf9, Oas1a and Isg15 in an IFNAR dependent manner, with levels mirroring that observed in myeloid cells (Fig. 1a-b; Supplementary

Fig. 1a-b). Further, as in myeloid cells8,25, IFNb treatment significantly enhanced adipocyte IFNAR-dependent, LPS-driven proinflammatory cytokine production (Fig. 1c,

Supplementary Fig. 1c). Priming of adipocytes was not restricted to IFNb as an IFNα subtype (e.g., IFNα4) similarly enhanced LPS-driven IL-6 production (Supplementary

Fig. 1d). In addition to Toll-like receptor 4 (TLR4), primary adipocytes also expressed

TLR2, TLR3, and TLR9 (Supplementary Fig. 1e), while activation of TLR2 (Pam2Cys) or TLR3 (Poly I:C) signaling in adipocytes was sufficient to induce IL-6, IFNb production and activate type I IFN axis (Supplementary Fig. 1f-i). Overall these findings suggest that akin to myeloid cells, various TLR ligands can potently induce proinflammatory cytokine production and activate the type I IFN axis in adipocytes. Further, our data suggest that activation of the type I IFN/IFNAR axis regulates adipocyte inflammatory vigor.

This similar type I IFN tuning of inflammatory vigor between adipocytes and myeloid cells (e.g., macrophages) prompted us to investigate the breadth of their shared functionality. Utilizing an unbiased RNA-seq approach (Fig. 1d), principal component analysis revealed that despite clear distinction at baseline between adipocytes and macrophages, treatment with IFNb followed by LPS drove adipocytes and macrophages to converge towards a similar gene expression signature (Fig. 1e-f). At baseline, only 35 out of the 2500 most highly expressed genes (0.7%; most overlapped genes are putative) in adipocytes and myeloid cells were shared, indicating a comparison of distinct cell types

(Fig. 1g; Supplementary Fig. 2a). Of note, while IFNb or LPS alone increased the overlap of gene expression patterns (1206 [25.1%], 758 [20.2%] respectively) between the cell types, IFNb/LPS combined treatment led to the greatest convergence of regulated genes (2033 [30.4%]; Fig. 1g; Supplementary Fig. 2b). Comparably upregulated genes between the adipocytes and macrophages, after combined IFNb/LPS treatment, included those associated with broad activation of inflammatory cascades and antigen presentation (Fig. 1h; Supplementary Fig. 2c). Notably, 42% of genes were overlapped between IFNb+LPS and IFNb alone, while only 8.3% were overlapped between

IFNb+LPS and LPS alone in adipocytes (Supplementary Fig. 2d). Closer examination of pathways significantly enhanced by combined IFNb/LPS (>2 fold over IFNb or LPS alone) revealed ontologies associated with diabetic complications, TNF and IL-6 signaling, STAT3 pathway, secreted soluble factors (e.g., IL-15, CXCL9, CXCL10, CCL5),

PI3K/AKT activation, impaired glucose tolerance, cytokine signaling, and TLR signaling in adipocytes (Fig. 1i). In addition, complementary computational analyses of enriched transcription factor (TF) binding site ‘motifs’ highlighted a convergence between

IFNb/LPS stimulated adipocytes and macrophages (e.g. STATs, IRFs, NF-Kb, chromatin accessibility; Supplementary Fig. 3a-b). Collectively these findings suggest that adipocytes possess a dormant underlying immunological capacity similar to bone marrow derived cells of myeloid origin and that the activation of the type I IFN/IFNAR axis in adipocytes is, in part, responsible for uncovering adipocyte inflammatory clades.

Alteration of glycolysis and fatty acid oxidation regulates myeloid cell inflammatory capacity26-28 and glucose metabolism mediates IFNβ antiviral capacity29. Our unbiased

RNA-seq analysis of IFNb treated adipocytes also revealed augmented expression of genes regulating glycolytic pathways (e.g. HIF1A30, EIF631; Fig. 2a). To examine the direct effect of IFNβ on adipocyte metabolic flux through glycolysis, the expression of key glycolysis-associated enzymes, lactate production, and cellular respiration was quantified. IFNb treatment of adipocytes enhanced phosphofructose kinase (Pfk1), phosphoglycerate kinase (Pgk1) and pyruvate kinase (Pkm2) mRNA expression, promoted basal extracellular acidification (ECAR), without altering lactate production, and increased cellular respiration (OCR) (Fig. 2c-d; Supplementary Fig. 4a-c). Combined

IFNb+LPS did not further enhance OCR or ECAR, suggesting effects are predominantly

IFNβ-driven (Supplementary Fig. 4d-e). These data were indicative that IFNβ altered aerobic glycolysis in adipocytes. To begin to delineate metabolic pathways modulated by

IFNβ and to examine if such alterations are linked to glycolysis, adipocytes were treated with IFNβ in the presence of inhibitors of fatty acid oxidation (etomoxir) or glycolysis (2-

Deoxy-D-Glucose [2-DG]). Only 2-DG treatment was sufficient to reverse the IFNb- mediated increase in basal cellular respiration (Fig. 2d; Supplementary Fig. 4f), dampen the expression of IFNb-driven signature genes (Oas1a and Isg15, Fig. 2e) and depress

IFNb-driven, IFNAR-dependent, augmentation of adipocyte inflammatory potential (Fig.

2f; Supplementary Fig. 4g). Combined, these findings suggest that the type I IFN axis alters adipocyte core metabolism, possibly through aerobic glycolysis, and that such modification is associated with the enhancement of adipocyte inflammatory potential.

Obesity is linked with augmented levels of various systemic TLR triggers (e.g.,

LPS22, DNA32) known to induce type I IFN production. WT mice fed a high fat diet (HFD), compared to chow diet (CD), had an enhanced expression of type I IFN signature genes

Oas1a and Isg15 in spleen, liver, and various fat depots (iWAT, eWAT, pWAT)

(Supplementary Fig. 5). As adipocytes comprise the core of WAT, expression and activation of type I IFN axis in adipocytes was examined next. Primary adipocytes from

HFD-fed WT mice, compared to CD-fed controls, displayed an augmented type I IFN signature including Ifnb1, Ifnar1, Oas1a, and Isg15 (Fig 3a). Further, in an IFNAR- dependent manner, IFNb primed adipocytes from HFD mice, compared to CD-fed controls, were significantly more vigorous in their IL-6 output after LPS challenge (Fig.

3b).

Pathological accumulation of WAT/adipocytes and inflammation are hallmarks of obesity development and pathogenesis of obesity-associated sequelae. IFNAR signaling modulates inflammatory potential of both immune cells and adipocytes. HFD fed IFNAR-

/- and WT mice exhibited similar obesity development as assessed by lack of differential total body weight (Supplementary Fig. 6a; in agreement with a recent report9), energy expenditure, food intake, systemic cholesterol, total body adiposity, BAT UCP-1 expression and adipocyte gross morphological appearance (Supplementary Fig. 6b-g).

However, genetic modulation of IFNAR signaling altered WAT distribution (decreased inguinal and perirenal WAT mass, and increased eWAT mass; Supplementary Fig. 7a- c). Progressive decline in eWAT is associated with increased adipocyte death, enhanced inflammation and total body insulin resistance33,33. Despite increased eWAT size, lack of

IFNAR signaling correlated with reduced total immune cell infiltration (CD45+) and specifically with diminished numbers of infiltrating T cells (CD3+CD4+, CD3+CD8+) and B cells (CD11c-B220+) and expression of the T and B cell chemoattractants (T cells: Cxcl10,

Cxcl9; B cells: Cxcl13 and Ltb4r) (Fig. 3c; Supplementary Fig. 8; Supplementary Table

1). Although the total number of eWAT infiltrating macrophages (F4/80+CD11b+) and neutrophils (GR1hiCD11b+; data not shown [DNS]) were similar between WT and IFNAR-

/- mice, lack of IFNAR signaling significantly reduced macrophage proinflammatory cytokine production (IL-6 and TNF; Fig. 3c; Supplementary Table 1) – cytokines central to obesity pathogenesis34.

In agreement with reduced inflammatory setting, compared to their age-, diet- and weight-matched WT controls, obese IFNAR-/- mice exhibited improved glucose metabolism, as determined by glucose and insulin tolerance tests (Fig. 3d-e). Further,

63 despite similar hepatic steatosis (cholesterol and triglyceride levels) and liver morphology between obese IFNAR-/- and WT counterparts (Supplementary Fig. 9), obese IFNAR-/- mice had attenuated hepatocellular injury as measured by systemic alanine transaminase

(ALT; Fig. 3f) levels. Reduced hepatocellular damage in obese IFNAR-/- mice correlated with decreased total hepatic immune cell numbers and numbers of CD4+ T cells, but similar numbers of liver infiltrating CD8+ T cells and macrophages. However, as in eWAT, lack of IFNAR signaling reduced liver macrophage proinflammatory cytokine production

(IL-6 and TNF) and IFNg production by hepatic CD8+ T cells — the latter in agreement with a published report9 (Fig. 3g; Supplementary Table 2 and DNS). These findings suggest that obesity promotes activation of the type I IFN/IFNAR axis and that type I IFN sensing by IFNAR modulates inflammation and pathogenesis of obesity-associated sequelae independent of weight gain.

The contribution of hematopoietic or non-hematopoietic IFNAR expression to obesity pathogenesis, via reciprocal bone marrow transfers between WT and IFNAR-/- mice, was examined next (Fig. 3h). Successful reconstitution was confirmed by flow cytometry (Supplementary Fig. 10a). Both hematopoietic (KO to WT) and non- hematopoietic (WT to KO) locus of IFNAR expression impacted LPS-driven IL-6 and TNF production (Fig. 3i-j). Further, in context of HFD feeding, either locus of IFNAR expression (hematopoietic or non-hematopoietic) equally impacted total body weight gain, adipose tissue mass (eWAT, iWAT and pWAT) and liver weight (Supplementary

Fig. 10b-f). Such parallels between hematopoietic and non-hematopoietic IFNAR expression largely correlated with similar obesity associated immune cell infiltration into eWAT (e.g., increase in total CD3+CD4+ cells, decreases in CD3+CD8+,

F4/80+CD11b+TNF+, F4/80+CD11b+IL-6+ cells) and liver (e.g., increases in CD3+CD4+ and CD3+CD8+ cells, decreases in F4/80+CD11b+TNF+, F4/80+CD11b+IL-6+ cells), and mirrored severity of glucose dysmetabolism and hepatocellular damage (ALT 2 to 3-fold less) (Fig. 3k-n; Supplementary Tables 3-4). These findings suggest that non- hematopoietic IFNAR expression is a relevant contributor to obesity-associated inflammation and pathogenesis of obesity-associated sequelae.

The potential role of adipocyte IFNb sensing to hepatocyte injury was examined next. IFNb sensing by HepG2 hepatocytes in the presence of LPS did not result in augmented hepatocellular injury, as quantified by ALT release (Fig. 3o left). However, culturing of HepG2 hepatocytes with conditioned media (CM) from mouse primary adipocytes exposed to IFNb and LPS exacerbated hepatocellular injury (Fig. 3o right).

Adipocyte-associated enhancement of hepatocellular injury correlated with increased IL-

6 and TNF levels within the CM – direct mediators of hepatocyte death35 (Supplementary

Fig. 11). These findings suggest that IFNb-driven production of adipocyte mediators are sufficient to exacerbate hepatocyte injury. Collectively, our data indicate proportional contribution of hematopoietic or non-hematopoietic IFNAR activation to the severity of obesity-associated metabolic derangements. Further, our findings imply that adipocytes type I IFN sensing may be an important modulator of such effects.

To determine whether effects of type I IFN/IFNAR axis activation are conserved from mice to humans, primary adipocytes from persons undergoing bariatric procedures were examined. Adipoq and FABP4 mRNA expression was used to determine the efficacy of adipocyte differentiation (Supplementary Fig. 12). LPS treatment of primary human adipocytes was sufficient to augment IFNb production and to induce the mRNA

65 expression of the type I IFN axis and including IRF1, OAS1 and ISG15 (Fig. 4a-b). In addition, similar to mice, IFNb treatment significantly increased the capacity of human adipocytes to produce LPS-driven IL-6 (Fig. 4c), while treatment with 2-DG was sufficient to reverse augmented IL-6 production (Fig. 4d).

Given that the type I IFN/IFNAR axis modifies NAFLD pathogenesis in mice, we next probed whether IFNb responsiveness in human adipocytes is altered in the presence of obesity-associated hepatocellular disease. Persons with severe obesity were stratified into either metabolically healthy (Met-H) or metabolically challenged (Met-C) groups according to well-established clinical parameters including markers of hepatocellular disease (i.e., NAFLD activity score (NAS), aspartate transaminase [AST], ALT, gamma- glutamyltransferase [GGT]; Fig. 4 e-g, Supplementary Table 5). Despite similar BMI and systemic lipid profiles (i.e., total cholesterol, low-density lipoprotein [LDL], high-density lipoprotein [HDL], and serum triglycerides; Supplementary Table 5) between the two cohorts, IFNb alone was sufficient to promote IL-6 production in Met-C adipocytes as compared to Met-H adipocytes. Enhancement of IL-6 production was even greater in LPS treated cells, while the combination of IFNb/LPS treated adipocytes led to the highest IL-

6 production (Fig. 4h). As the type I IFN/IFNAR axis activates a multitude of inflammatory signaling hubs, we further examined whether a type I IFN signature36 is associated with elevated signs of hepatocellular disease (NAS, AST, ALT, and GGT levels). In fact, Met-

C individuals had heightened systemic levels of type I IFN signature cytokines36 including

IFNb, TNF, CCL2 and a trending increase in IL-6, CCL3, CXCL9 and CXCL10 as compared to Met-H individuals (Fig. 4i; Supplementary Table 6). Furthermore, systemic

IFNβ levels were directly correlated with hepatocellular disease (Figure 4j). Collectively,

66 these results demonstrate that the type I IFN axis and its effects on inflammatory capacity are conserved in human adipocytes. Further, our data suggest that detection of the type

I IFN/IFNAR axis-associated signatures positively correlated with obesity-driven hepatocellular damage in humans.

Together, our findings underscore a previously unappreciated role of type I

IFNs/IFNAR axis in the regulation of adipocyte inflammatory vigor. Although partial similarity in inflammatory capabilities between adipocytes and myeloid cells, including proinflammatory cytokines3, expression of innate immune receptors4, MHC class I and class II molecules5-7, have been previously hinted at, similarity of their gene expression patterns, especially in context of type I IFN sensing, had not been elucidated. The potential for the type I IFN axis to unlock a fundamental functional convergence between inflammatory gene expression patterns in adipocytes and macrophages emphasizes the complexity of white adipose tissue biology. In mice, type I IFNs are composed of 14 different IFNα subtypes and IFNω, IFNε, IFNτ, IFNκ and IFNb. Activation of pattern- recognition receptors results in an anatomical locus/tissue/cell type specific37 production of a fraction of IFNα subtypes and IFNb. However, IFNβ is specifically induced by LPS stimulation in myeloid cells38 and holds a unique ability to specifically interact with IFNAR1 in an IFNAR2-independent manner (in addition to its ability to interact with the IFNAR1 and 2 heterodimeric complex39. Although our data indicate the IFNα4 subunit can also potentiate adipocyte inflammatory vigor, formal definition of individual IFNα subtypes sufficient for uncovering adipocyte inflammatory capacity would be extremely challenging.

In addition, as adipocytes and immune cells (e.g., dendritic cells, macrophages) within

67 the adipose tissue can produce type I IFNs, the source of type I IFNs in the context of obesity warrants exploration.

Our additional observations suggest, to our knowledge for the first time, that the type I IFN axis alters adipocyte core metabolism, possibly through aerobic glycolysis as our data suggest that IFNb enhances basal OCR and ECAR. Although lactate is a contributor to ECAR levels, other sources, including CO2 from the citric acid cycle40, likewise can contribute to acidification. Our findings additionally suggest the possibility that IFNb modification of adipocyte core metabolism and inflammatory vigor are interlinked. IFNb regulates glucose uptake and metabolism in cells28,29, and IFNb driven glucose metabolism is an important mechanism for the induction of antiviral responses29.

Notably, type I IFNs are capable of altering human macrophage epigenomes to broadly reprogram their LPS-driven responses41 and modification of core cellular metabolism alters epigenetic and transcriptional networks associated with augmented inflammatory capacity (e.g., naïve CD4+ T cell polarization, IFNg, NF-kB signaling)42 in immune cells — potentially invoking another possible similarity between adipocytes and immune counterparts.

Extensive literature has long indicated that the type I IFN/IFNAR axis is a promoter of hepatic and metabolic dysfunction9-20. However, recent evidence has suggested that

FABP4cre-driven IFNAR deletion may modify HFD-driven obesity and downstream metabolic sequelae21. However, FABP4cre is not restricted to adipocytes (e.g., adipocyte precursor cells, macrophages, endothelial cells, osteogenic cells, ganglion, adrenal medulla, and liver)21,43,44. In agreement with this existing literature, this study likewise demonstrates incomplete deletion of IFNAR1 in adipose tissue and thus it is difficult to

68 draw strong conclusions. A positive correlation between hepatic type I IFN signature and

TNF expression, a well-established inducer of hepatocyte death/disease, in adult bariatric patients was identified. Additionally, enhanced expression of type I IFN axis genes (i.e.,

IFIT1, MX1, OAS1)21 was observed in adipose tissue from bariatric patients prior to surgery, findings in line with our observations. Given our findings, a more specific adipocyte-targeted in vivo approach utilizing Adipoqcre mice to closely examine the impact of the IFNAR axis is warranted and could yield novel appreciation for adipocyte-specific contribution to obesity-associated inflammation.

It has been posited that obesity-associated metabolic status is a consequence of a balance in “healthy” WAT45. Although inflammation is considered a driver of the progression to dysfunctional WAT and a consequent metabolically challenged state, mechanisms underlying such processes have not been elucidated. In addition, our pediatric cohort represents a unique group and gives us a keen opportunity to examine the development/progression of NASH from a young age – mostly devoid of the accumulation of insults driving adult NASH. As such, our findings indicate enhancement of the type I IFN axis and its effects on adipocytes in pediatric bariatric patients. IFNb therapy, an approach for clinical care of multiple sclerosis patients, impairs glucose tolerance and insulin sensitivity18 and is associated with hepatic dysfunction19,20. Whether the type I IFN axis effects on adipocytes modulates the transition between healthy and dysfunctional WAT remains to be determined and further exploration is fully warranted.

Thus, future definition of the similarities and differences between pediatric and adult obese patients would be of significant interest.

Prior dogma has proposed that adipose tissue is a relative immunological sanctuary. However, given its high energy source, there is a propensity for certain microbes, including bacteria46, viruses47, and parasites28 to favor adipose tissue/adipocytes. In addition, the obesity-associated state predisposes to poor outcomes to viral (e.g. Influenza) and bacterial (e.g. sepsis) infections48, established inducers of type I IFNs. It is reasonable to posit that these systemic infections, via potentiation of systemic type I IFN-driven responses, may enhance adipocyte inflammatory vigor and exacerbate infection-associated inflammation-driven sequelae. In fact, our data indicates that systemic IFNb was sufficient to trigger WAT IFNB1 expression – suggestive that IFNb produced in other cells/organs can be sensed and responded to by WAT. Hence, investigations into whether utilization of available biological inhibitors of IFNAR49 to dampen systemic infection triggered intra-adipocyte inflammation and promote improved outcomes would be interest.

In sum, our report highlights a potentially undervalued role for adipocytes in the context of type I IFN-driven pathogenic diseases and suggest that greater examination of these cells could lead to possible novel insights into disease pathogenesis. In fact, it is plausible that novel pharmacological intervention into the type I IFN/IFNAR axis function, in both adipocytes and immune cells, would provide novel approaches to dampen type I

IFN-driven diseases, including obesity-associated metabolic harm.

ACKNOWLEDGEMENTS

This study was supported, in part, by CCHMC Pediatric Diabetes and Obesity Center (to

S.D.), NIH R01DK099222 (to S.D); R01DK099222-02S1 (associated with S.D. and

M.E.M.F); AHA 17POST33650045 (to M.E.M.F); NIH T32AI118697 and T32GM063483-

14 (associated with C.C.C.); and PHS Grant P30 DK078392 Pathology of the Digestive

Disease Research Core Center at CCHMC (associated with S.D.). We thank Matthew

Lawson, Adrienne Wilburn, Vishakha Sharma, Jarren Oates and the CCHMC Pediatrics

Diabetes and Obesity Center team for technical assistance.

AUTHOR CONTRIBUTIONS

C.C.C., M.C., M.E.M.-F., T.E.S., M.S.M.A.D, P.C.A., R.M., and M.W. participated in data generation. C.C.C., M.C., M.E.M.-F., T.E.S., M.S.M.A.D, P.C.A., R.M., M.W., M.A.H.,

T.H.I., S.D. participated in data analysis, interpretation, provided materials and technical support and participated in review of the manuscript. C.C.C., M.E.M.F and S.D. obtained the funding. C.C.C. and S.D. participated in the conception and design of the study and wrote the manuscript.

COMPETING FINANCIAL INTERESTS

The authors declare no competing financial interests

REFERENCES

1. Reilly, S.M. & Saltiel, A.R. Adapting to obesity with adipose tissue inflammation. Nat Rev Endocrinol 13, 633-643 (2017). 2. Weisberg, S.P., et al. Obesity is associated with macrophage accumulation in adipose tissue. J Clin Invest 112, 1796-1808 (2003). 3. Sewter, C.P., Digby, J.E., Blows, F., Prins, J. & O'Rahilly, S. Regulation of tumour necrosis factor-alpha release from human adipose tissue in vitro. J Endocrinol 163, 33-38 (1999). 4. Schaffler, A. & Scholmerich, J. Innate immunity and adipose tissue biology. Trends Immunol 31, 228-235 (2010). 5. Deng, T., et al. Class II major histocompatibility complex plays an essential role in obesity-induced adipose inflammation. Cell Metab 17, 411-422 (2013). 6. Xiao, L., et al. Large adipocytes function as antigen-presenting cells to activate CD4(+) T cells via upregulating MHCII in obesity. Int J Obes (Lond) 40, 112-120 (2016). 7. Malide, D., Yewdell, J.W., Bennink, J.R. & Cushman, S.W. The export of major histocompatibility complex class I molecules from the endoplasmic reticulum of rat brown adipose cells is acutely stimulated by insulin. Mol Biol Cell 12, 101-114 (2001). 8. Cappelletti, M., et al. Type I interferons regulate susceptibility to inflammation- induced preterm birth. JCI Insight 2, e91288 (2017). 9. Ghazarian, M., et al. Type I Interferon Responses Drive Intrahepatic T cells to Promote Metabolic Syndrome. Sci Immunol 2(2017). 10. Wang, X.A., et al. Interferon regulatory factor 7 deficiency prevents diet-induced obesity and insulin resistance. Am J Physiol Endocrinol Metab 305, E485-495 (2013). 11. Wang, X.A., et al. Interferon regulatory factor 3 constrains IKKbeta/NF-kappaB signaling to alleviate hepatic steatosis and insulin resistance. Hepatology 59, 870- 885 (2014). 12. Kumari, M., et al. IRF3 promotes adipose inflammation and insulin resistance and represses browning. J Clin Invest 126, 2839-2854 (2016). 13. Kissig, M., et al. PRDM16 represses the type I interferon response in adipocytes to promote mitochondrial and thermogenic programing. EMBO J 36, 1528-1542 (2017). 14. Franceschini, L., et al. Reciprocal interference between insulin and interferon- alpha signaling in hepatic cells: a vicious circle of clinical significance? Hepatology 54, 484-494 (2011). 15. Bhattacharya, A., et al. Superoxide Dismutase 1 Protects Hepatocytes from Type I Interferon-Driven Oxidative Damage. Immunity 43, 974-986 (2015). 16. York, A.G., et al. Limiting Cholesterol Biosynthetic Flux Spontaneously Engages Type I IFN Signaling. Cell 163, 1716-1729 (2015). 17. Hannibal, T.D., et al. Deficiency in plasmacytoid dendritic cells and type I interferon signalling prevents diet-induced obesity and insulin resistance in mice. Diabetologia 60, 2033-2041 (2017).

18. Koivisto, V.A., Pelkonen, R. & Cantell, K. Effect of interferon on glucose tolerance and insulin sensitivity. Diabetes 38, 641-647 (1989). 19. Tremlett, H.L., Yoshida, E.M. & Oger, J. Liver injury associated with the beta- interferons for MS: a comparison between the three products. Neurology 62, 628- 631 (2004). 20. Francis, G.S., et al. Hepatic reactions during treatment of multiple sclerosis with interferon-beta-1a: incidence and clinical significance. Drug Saf 26, 815-827 (2003). 21. Wieser, V., et al. Adipose type I interferon signalling protects against metabolic dysfunction. Gut (2016). 22. Cani, P.D., et al. Metabolic endotoxemia initiates obesity and insulin resistance. Diabetes 56, 1761-1772 (2007). 23. Uematsu, S. & Akira, S. Toll-like receptors and Type I interferons. J Biol Chem 282, 15319-15323 (2007). 24. Ivashkiv, L.B. & Donlin, L.T. Regulation of type I interferon responses. Nat Rev Immunol 14, 36-49 (2014). 25. Cappelletti, M., Lawson, M.J., Chan, C.C., Wilburn, A.N. & Divanovic, S. Differential outcomes of TLR2 engagement in inflammation-induced preterm birth. J Leukoc Biol (2017). 26. Ganeshan, K. & Chawla, A. Metabolic regulation of immune responses. Annu Rev Immunol 32, 609-634 (2014). 27. Pearce, E.L. & Pearce, E.J. Metabolic pathways in immune cell activation and quiescence. Immunity 38, 633-643 (2013). 28. Wu, D., et al. Type 1 Interferons Induce Changes in Core Metabolism that Are Critical for Immune Function. Immunity 44, 1325-1336 (2016). 29. Burke, J.D., Platanias, L.C. & Fish, E.N. Beta interferon regulation of glucose metabolism is PI3K/Akt dependent and important for antiviral activity against coxsackievirus B3. J Virol 88, 3485-3495 (2014). 30. Cheng, S.C., et al. mTOR- and HIF-1alpha-mediated aerobic glycolysis as metabolic basis for trained immunity. Science 345, 1250684 (2014). 31. Brina, D., et al. eIF6 coordinates insulin sensitivity and lipid metabolism by coupling translation to transcription. Nat Commun 6, 8261 (2015). 32. Nishimoto, S., et al. Obesity-induced DNA released from adipocytes stimulates chronic adipose tissue inflammation and insulin resistance. Sci Adv 2, e1501332 (2016). 33. Strissel, K.J., et al. Adipocyte death, adipose tissue remodeling, and obesity complications. Diabetes 56, 2910-2918 (2007). 34. Lackey, D.E. & Olefsky, J.M. Regulation of metabolism by the innate immune system. Nat Rev Endocrinol 12, 15-28 (2016). 35. Luedde, T., Kaplowitz, N. & Schwabe, R.F. Cell death and cell death responses in liver disease: mechanisms and clinical relevance. Gastroenterology 147, 765-783 e764 (2014). 36. Bauer, J.W., et al. Elevated serum levels of interferon-regulated chemokines are biomarkers for active human systemic lupus erythematosus. PLoS Med 3, e491 (2006). 37. Trinchieri, G. Type I interferon: friend or foe? J Exp Med 207, 2053-2063 (2010).

38. Gessani, S., Belardelli, F., Pecorelli, A., Puddu, P. & Baglioni, C. Bacterial lipopolysaccharide and gamma interferon induce transcription of beta interferon mRNA and interferon secretion in murine macrophages. Journal of virology 63, 2785-2789 (1989). 39. de Weerd, N.A., et al. Structural basis of a unique interferon-beta signaling axis mediated via the receptor IFNAR1. Nat Immunol (2013). 40. TeSlaa, T. & Teitell, M.A. Techniques to monitor glycolysis. Methods Enzymol 542, 91-114 (2014). 41. Park, S.H., et al. Type I interferons and the cytokine TNF cooperatively reprogram the macrophage epigenome to promote inflammatory activation. Nat Immunol 18, 1104-1116 (2017). 42. Raghuraman, S., Donkin, I., Versteyhe, S., Barres, R. & Simar, D. The Emerging Role of Epigenetics in Inflammation and Immunometabolism. Trends Endocrinol Metab 27, 782-795 (2016). 43. Martens, K., Bottelbergs, A. & Baes, M. Ectopic recombination in the central and peripheral nervous system by aP2/FABP4-Cre mice: implications for metabolism research. FEBS Lett 584, 1054-1058 (2010). 44. Lee, K.Y., et al. Lessons on conditional gene targeting in mouse adipose tissue. Diabetes 62, 864-874 (2013). 45. Crewe, C., An, Y.A. & Scherer, P.E. The ominous triad of adipose tissue dysfunction: inflammation, fibrosis, and impaired angiogenesis. J Clin Invest 127, 74-82 (2017). 46. Ozuru, R., et al. Adipose tissue is the first colonization site of Leptospira interrogans in subcutaneously infected hamsters. PLoS One 12, e0172973 (2017). 47. Damouche, A., et al. Adipose Tissue Is a Neglected Viral Reservoir and an Inflammatory Site during Chronic HIV and SIV Infection. PLoS Pathog 11, e1005153 (2015). 48. Huttunen, R. & Syrjanen, J. Obesity and the risk and outcome of infection. Int J Obes (Lond) 37, 333-340 (2013). 49. Furie, R., et al. Anifrolumab, an Anti-Interferon-alpha Receptor Monoclonal Antibody, in Moderate-to-Severe Systemic Lupus Erythematosus. Arthritis Rheumatol 69, 376-386 (2017). 50. Castaneda, T.R., et al. Metabolic control by S6 kinases depends on dietary lipids. PLoS One 7, e32631 (2012). 51. Harley, I.T., et al. IL-17 signaling accelerates the progression of nonalcoholic fatty liver disease in mice. Hepatology 59, 1830-1839 (2014). 52. Fischer, K., et al. Alternatively activated macrophages do not synthesize catecholamines or contribute to adipose tissue adaptive thermogenesis. Nat Med 23, 623-630 (2017). 53. O'Rourke, R.W., Gaston, G.D., Meyer, K.A., White, A.E. & Marks, D.L. Adipose tissue NK cells manifest an activated phenotype in human obesity. Metabolism 62, 1557-1561 (2013). 54. Giles, D.A., et al. Thermoneutral housing exacerbates nonalcoholic fatty liver disease in mice and allows for sex-independent disease modeling. Nat Med 23, 829-838 (2017).

55. Chen, J., Bardes, E.E., Aronow, B.J. & Jegga, A.G. ToppGene Suite for gene list enrichment analysis and candidate gene prioritization. Nucleic Acids Res 37, W305-311 (2009). 56. Muers, M. Functional genomics: the modENCODE guide to the genome. Nat Rev Genet 12, 80 (2011). 57. Portales-Casamar, E., et al. The PAZAR database of gene regulatory information coupled to the ORCA toolkit for the study of regulatory sequences. Nucleic Acids Res 37, D54-60 (2009). 58. Rosenbloom, K.R., et al. The UCSC Genome Browser database: 2015 update. Nucleic Acids Res 43, D670-681 (2015). 59. Heinz, S., et al. Simple combinations of lineage-determining transcription factors prime cis-regulatory elements required for macrophage and B cell identities. Mol Cell 38, 576-589 (2010). 60. Weirauch, M.T., et al. Determination and inference of eukaryotic transcription factor sequence specificity. Cell 158, 1431-1443 (2014).

Figure 1. IFNβ/IFNAR axis exacerbates adipocyte immune potential. Primary adipocytes isolated from CD fed WT and IFNAR-/- mice were treated with saline (NS),

IFNb (200 U/ml) or LPS (100 ng/ml) as indicated. (a) IFNb protein. (b) Expression of indicated type I IFN axis genes, relative expression to WT NS. (c) IL-6 protein levels in supernatant; % change over NS. (d-i) Treated adipocytes and macrophages subjected to

RNA-seq analysis. (d) Schematic overview. (e) Principal coordinate analysis, distance of component 2 (X-axis) delineates the similarity of gene expression patterns between cell types. (f) Heat maps of adipocytes and macrophages. (g) percent of shared genes between adipocytes and macrophages. (h) Venn diagram and volcano plots of genes expressed under IFNb+LPS treatment in adipocytes and macrophages. (i) Adipocyte ontology pathways with IFNb+LPS treatment (>2 fold over IFNb or LPS alone). (a-c)

Representative of 3 independent experiments, n = 3/condition. (d-i) A single experiment, n = 2/condition. Data represents mean +/- SEM. (a-c) Student’s t-test. *P < 0.05, **P <

0.01, ***P < 0.001, ****P < 0.0001. A denotes adipocytes. M denotes macrophages.

Figure 2. IFNβ modifies glycolysis-associated inflammatory vigor in adipocytes. (a)

Expression of representative genes in pathways associated with regulation of glycolysis in adipocytes. (b-g) Adipocytes treated in the presence or absence of IFNb (200 U/ml).

(b) Expression of indicated glycolysis genes, expression relative to NS. (c-d) Adipocyte cellular bioenergetics (oligomycin [10mM], FCCP [10mM]). (c) Basal ECAR. (d) Basal

OCR in the presence or absence of in the presence or absence of 2-Deoxy-D-glucose [2-

DG; 2mM]. (e) IFNb treated adipocyte expression of indicated type I IFN axis genes, relative expression to ctrl. (f) IL-6 protein levels in supernatant; % change over NS. (a) A

76 single experiment, n = 2/condition. (b-e) Representative of 3 independent experiments, n

= 3-6/condition. (f) Data combined from 2 independent experiments, n = 2-3/condition.

Data represents mean +/- SEM. (b-f) Student’s t-test. *P < 0.05, **P < 0.01, ***P < 0.001,

****P < 0.0001.

Figure 3. Non-hematopoietic IFNAR axis strongly contributes to the pathogenesis of obesity-associated sequelae. (a) Expression of the indicated type I IFN axis genes in primary adipocytes isolated from WAT of WT mice fed a chow diet (CD) or high-fat diet

(HFD) for 8 weeks, relative expression to CD. (b) Primary adipocytes treated with saline

(NS), IFNb (200 U/ml) or LPS (100 ng/ml) as indicated and IL-6 protein levels in supernatant were quantified; % change over NS. (c-g) WT and IFNAR-/- mice were fed a

HFD for 22 weeks. (c) Average eWAT immune cell infiltration. (d) Glucose and (e) Insulin tolerance tests. (f) Alanine Transaminase. (g) Average liver immune cell infiltration. (h-n)

Reciprocal bone marrow transfers (BMT) between WT and IFNAR-/- mice were performed and successful reconstitution was confirmed at d74 post-transfer. (h) Schematic overview. In vivo cytokine capture assay (IVCCA) LPS-driven (i) IL-6 and (j) TNF levels in lean BMT mice. (k-n) Reciprocal BMT mice were placed on a HFD for 18 weeks. (k)

Average eWAT immune cell infiltration; % change over WT. (l) Glucose tolerance test.

(m) Liver immune cell infiltration; % change over WT. (n) Alanine transaminase (ALT).

(o) ALT of HepG2 cells cultured in the presence or absence of adipocyte conditioned media. (a-b) Representative of 2 independent experiments, n = 3/condition. (c-g)

Representative of 3 independent experiments, n = 6-7/condition. (h-n) A single experiment, n = 4-5/condition. (o) Data combined from 3 independent experiments, n =

2-3/condition. Data represents mean +/- SEM. (a-c, f-g, i-k, m-o) Student’s t-test. (d-e, i)

Area under curve. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

Figure 4. Type I IFN axis effects on adipocyte inflammatory vigor are conserved in humans. (a-d) Human primary adipocytes isolated from bariatric patients. Adipocytes were treated with Saline (NS), human IFNb (250U/ml) or LPS (100ng/ml) as indicated. (a)

Quantified supernatant IFNb protein. (b) Expression of indicated type I IFN axis genes, relative expression to NS. (c) IL-6 protein levels in supernatant; % change over NS. (d)

IL-6 protein levels in supernatant in the presence or absence of 2-DG (2μM), % change over NS. (e-j) Persons with severe obesity were stratified into Metabolically healthy (Met-

H) or Metabolically challenged (Met-C) groups. (e) NAFLD activity score. (f) ALT. (g) AST.

(h) IL-6 protein levels in adipocyte supernatant, % change over NS. (i) Average systemic protein levels of indicated cytokines and chemokines. (j) IFNβ correlation with AST. (a-d)

Representative patients, n = 3-5/condition. (e-j) Data combined, n = 11 Met-H and n = 12

Met-C. Data represents mean +/- SEM. (a-i) Student’s t-test. (j) Linear regression analysis. *P < 0.05, **P < 0.01, ****P < 0.0001.

ONLINE METHODS

Mouse Obesogenic Diet Model

All mice used were males on a C57BL/6 background (Jackson). WT and IFNAR-/- (bred in house8), mice were bred at Cincinnati Children’s Hospital Medical Center (CCHMC) in a specific pathogen-free (spf) facility maintained at 22˚C, with free access to autoclaved low-fat chow diet food (LAB Diet #5010; calories provided by carbohydrates [58%], fat

[13%] and protein [29%]) and water. At 6-8 weeks of age, mice were fed either an irradiated high-fat diet (HFD; Research Diets #D12492; 60% of calories from fat) or a CD.

Food was replenished on a weekly basis to avoid contamination. Total body fat, lean and water mass were determined by nuclear magnetic resonance (Whole Body Composition

Analyzer; Echo MRI)50. Mice were fasted overnight prior to glucose metabolism testing, insulin tolerance testing, or terminal harvest. Glucose and insulin tolerance tests (ITT) were done as previously described51. Briefly, mice were fasted overnight and glucose tolerance was determined by injecting mice with 100 µL of a 10% dextrose solution per gram of body weight. Glucose levels were kinetically quantified at the times indicated. For

ITT, mice received 100 µL of a 0.15 units/mL solution of insulin (Novolin) per gram of body weight. Energy expenditure measured by Phenomaster (TSE systems). All care was provided in accordance with the Guide for the Care and Use of Laboratory Animals. All studies were approved by the CCHMC IACUC.

Mouse Primary Adipocytes

Inguinal white adipose tissue was isolated and digested (1mg/ml Collagenase Type IV,

52 Dispase 2, CaCl2) as previously described . Stromal vascular fraction containing

83 predadipocytes was cultured until confluence. Preadipocytes were differentiated as previously described52. Briefly, preadipocytes were subjected to initiation media (Growth media [DMEM:F12, FBS, Pen/Strep], Rosiglitazone, Dexamethasone, 3-Isobutyl-1- methylxanthine, insulin) for 2 days. Afterwards, cells were switched to continuation media

(Growth media, Rosiglitazone, Insulin) for 2 days, followed by differentiation media

(Growth media, Insulin) for an additional 2 days. Differentiated adipocytes were utilized for downstream processes.

Human Subjects

Bariatric surgery participants were recruited from the Cincinnati Children’s Hospital

Medical Center (CCHMC) Pediatric Diabetes and Obesity Center. Patients with alcohol abuse, viral and autoimmune hepatitis, immunosuppressive or steroid use were excluded.

Liver sections were examined qualitatively using the pediatric NASH-CRN scoring system by a certified liver pathologist. The NAFLD activity score (NAS) is a sum of scores for steatosis, lobular inflammation and ballooning. Patients were segregated into a metabolically healthy (Met-H) or metabolically challenged (Met-C) category based on these well-established clinical parameters of hepatocellular disease (Met. H, n = 10; Met.

C, n = 18; clinical phenotypes provided in Supplementary Table 1). Recruitment and study protocols were approved by the institutional review board at CCHMC.

Human Primary Adipocytes

Omental white adipose tissue was collected at the time of surgery and processed as previously described protocols53. Briefly, adipose was minced and digested with Type II

84 collagenase (40mg/ml in PBS with 2% BSA) for 45 min at 37C. Digested tissues were filtered and centrifuged at 250g to isolate the stromal vascular fraction (SVF). SVF was subjected Ack lysis buffer. SVF was cultured (in Expansion media [DMEM/F:12, 15%

FBS, 1% Pen-strep]) until confluence and subjected to human adipocyte differentiation media (DMEM/F:12, 1% Pen-strep, 2mM glutamine, 15mM HEPES, 10mg/ml transferrin,

33µM biotin, 0.5µM insulin, 17µM pantothenate, 0.1µM dexamethasone, 2nM T3, 500µM

IBMX, 1µM ciglitazone) for 14-16d, followed by 7-10d in human adipocyte maintenance media (DMEM/F:12, 1% Pen-strep, 2mM glutamine, 15mM HEPES, 10mg/ml transferrin,

33µM biotin, 0.5µM insulin). Maximally differentiated adipocytes were utilized for downstream processes.

Reciprocal Bone Marrow Transfer and In vivo Cytokine Quantification

Reciprocal bone marrow transfers were generated using 8-week-old WT or IFNAR-/- recipient mice as previously described8. Briefly, 5 × 106 bone marrow cells were derived from the femurs of WT or IFNAR-/- mice and transferred into whole-body irradiated WT or

IFNAR-/- recipient mice. Peripheral blood chimerism was assessed by flow cytometric analysis 10 weeks after bone marrow reconstitution. In vivo cytokine capture assay was utilized to quantify systemic IL-6 and TNF levels were quantified using in vivo cytokine capture assay (IVCCA) as previously described8. Briefly, biotinylated capture antibodies against IL-6 (clone MP5-32C11) and TNF (clone TN3) (both eBioscience) were given via intraperitoneal injection 3 hours prior to LPS challenge and 4 hours later serum cytokine levels were determined.

Epididymal WAT (eWAT) immune infiltration isolation and analysis eWAT was isolated from obese WT and IFNAR-/- mice as previously described2. Isolated stromal vascular fraction cells were stimulated for 4h with PMA (50ng/ml; Sigma-Aldrich) and Ionomycin (1µg/ml; Calbiochem). Briefly, cells were stained with Live/dead stain

(Zombie UV Dye; Biolegend), B220 (clone RA3-6B2; Biolegend), CD45 (clone 104),

CD11b (clone M1/70), F4/80 (clone BM8), Gr1 (clone RB6-8C5), CD4 (clone GK1.5), CD8

(clone 53-6.7), IFNg (clone XMG1.2) TNF (clone MP6-XT22), and IL-6 clone (MP5-20F3)

[all ebioscience]. Data were collected using a LSR Fortessa flow cytometer (BD

Biosciences) and analyzed by FlowJo software (Tree Star).

Adipocyte cytokine quantification

Murine primary adipocytes were cultured in the presence or absence of IFNβ (200U/ml) for 3 hours and thereafter stimulated with saline, LPS (100ng/ml), Pam2 (100ng/ml) or

Poly(I:C) (25ug/ml) for 4 hours. IL-6 was quantified by ELISA (BD biosciences) as per manufacturer instructions.

Type I IFN Quantification

IFNβ levels in adipocyte culture supernatants was quantified with reference to a recombinant mouse IFN-β using an L-929 cell line transfected with an interferon-sensitive luciferase construct as previously described8. Luciferase activity was quantified on a

SpectraMax L luminometer (Molecular Devices).

qRT-PCR

Adipocytes were homogenized in TRIzol (Invitrogen) followed by RNA extraction, reverse transcription to cDNA (Verso cDNA Synthesis Kit; Thermo Scientific) and qPCR analysis

(Light Cycler 480 II; Roche) – according to manufacturer’s instruction as previously described54.

The following primers pairs were used for mouse studies: IFNB1 For

TCCAGCTCCAAGAAAGGACG Rev TTGAAGTCCGCCCTGTAGGT– IFNAR1 For

ACACTGCCCATTGACTCTCC Rev TTGGGTGCTACCCTCAGC – IRF9 For

ACAACTGAGGCCACCATTAGAGA Rev CACCACTCGGCCACCATAG – Oas1a For

AGCAGGTAGAGAACTCGCCA Rev CTGCATCAGGAGGTGGAGTT – ISG15 For

GTCACGGACACCAGGAAATC Rev AAGCAGCCAGCCGCAGACTG – IL-6 For

TGGTACTCCAGAAGACCAGAGG Rev AACGATGATGCACTTGCAGA – PFK1 For

CATGGGGAGAGAGGACAGA Rev AGTTCGGGAACAAGACGTTG – PGK1 For

CAGCCTTGATCCTTTGGTTG Rev CTGACTTTGGACAAGCTGGA – PKM1 For

AGCAGGTCCTTGGAAACCTT Rev AAGGAGTTTCATCTGGCCCT –Rev

TTGGAGTCAGCGCAGATCTG – CXCL10 For CCTATGGCCCTCATTCTCAC Rev

CGTCATTTTCTGCCTCATCC – Actb For GGCCCAGAGCAA GAGAGGTA Rev

GGTTGGCCTTAGGTTTCAGG.

The following primer pairs were used for human studies: hIRF1 For

CATGAGACCCTGGCTAGAGATG Rev TCCGGAACAAACAGGCATCC – hOas1 For

TGAGGTCCAGGCTCCACGCT Rev GCAGGTCGGTGCACTCCTCG – hISG15 For

GAGAGGCAGCGAACTCATCT Rev CTTCAGCTCTGACACCGACA – hIL-6 For

CATTTGTGGTTGGGTCAGG Rev AGTGAGGAACAAGCCAGAGC – hUBIQ For

CACTTGGTCCTGCGCTTGA Rev CAATTGGGAATGCAACAACTTTAT. mRNA

87 expression (in arbitrary units) of each gene was compared to Actb (beta-actin, mouse) expression or Ubiquitin (human).

Lactate Quantification Assay

Lactate levels in adipocyte culture supernatants were quantified by colorimetric assay kit

(Sigma-Aldrich) as per manufacturer instructions.

RNA sequencing and gene expression quantification

Gene expression of primary adipocytes and macrophages was determined as previously described by running 50 base pair single-end reads (~20 million reads per sample). All transcriptomic analyses were performed in StrandNGS. Following the removal of barcodes and primers, raw reads were aligned to the mm10 genome using annotations provided by UCSC with the following parameters: (1) minimum percent identify = 90; (2) maximum percent gaps = 5; (3) minimum aligned read length = 25; (4) number of matched to output per read = 1; and (5) ignore reads with more than 5 matches. The proprietary aligner (COBWeb) is based on the Burrow Wheeler Transform method. Aligned reads were used to compute reads per kilobase per million (RPKM) using the Expectation-

Maximization algorithm for the maximum likelihood estimation of expression. Further,

RPKM were thresholded at 1 and normalized using the DESeq algorithm, which computes a normalization factor (NF) for each sample. Within each sample, each transcript is divided by that transcript’s geometric mean across samples. The within-sample median of these values is that sample’s NF. To obtain normalized counts, a sample’s raw RPKM are divided by that sample’s NF. Finally, normalized per-transcript RPKM were baselined

88 to the median of all samples. Reasonably expressed transcripts (raw RPKM >3 in 100% of samples in at least one condition) were included for differential analysis. Differential expression was determined through 2-Way ANOVAs with an FDR-corrected p-value cutoff of 0.05 and a fold change requirement of >1.5. For pathway analysis, the database at toppgene.cchmc.org was employed, which amasses ontological data from over 30 individual repositories55. RNA-sequencing raw data can be accessed at GSE110236.

Transcription factor binding site enrichment analysis

To identify transcription factors (TFs) that might be regulating genes expressed in adipocytes or macrophages treated with IFNb (200U), LPS (100ng/ml) or IFNb/LPS, we utilized the RNA-sequencing analysis (above) coupled with a computational method that overlaps the genomic coordinates of a set of gene promoters with a large library of TF- genome interactions. We created the dataset library by compiling 510 mouse ChIP-seq and DNase-seq datasets from a variety of sources, including modENCODE56, PAZAR57, and the UCSC Genome Browser58. As input, our method takes a set of genomic regions of interest (e.g., promoters of genes whose expression changes upon stimulation), and systematically overlaps them with each ChIP-seq/DNase-seq dataset. The observed overlap of the input set with each dataset is calculated by counting the number of input regions it overlaps by at least one base. Next, a P-value describing the significance of this overlap is estimated using a simulation-based procedure. A distribution of expected overlap values is created from 2,000 iterations of randomly choosing RefSeq gene promoters with the same length as the input set (e.g., if 50 promoters of length 100 bp are used as input, then 50 randomly chosen promoters of length 100 bp will be used in

89 each simulation). The distribution of the expected overlap values from the randomized data resembles a normal distribution, and is used to generate a Z-score and P-value estimating the significance of the observed number of input regions that overlap each dataset. The resulting data thus provide ranked lists of datasets (TFs, histone marks, or open chromatin), based on experimentally-determined data located in the promoters of each gene set. We applied this procedure to each input gene list using three different promoter definitions: (-1000,+500), (-1000,+1000), and (-2000,+1), relative to the transcription start site. Results were similar regardless of promoter length. Additionally, we examined the underlying promoter sequences by calculating TF binding site motif enrichment scores (using the same promoter definitions). For this, we used the HOMER motif enrichment algorithm59, and a large library of mouse position weight matrices obtained from the CisBP database60. HOMER was run using the default null model.

Cellular Bioenergetics Quantification

Primary adipocytes were plated at 1x104 cells per well in a polyethylenimine pre-coated

XF96 Cell culture microplate. Prior to bioenergetics analysis, adipocytes were treated with saline, IFNb (200U/ml), etomoxir (250µM), and/or 2-DG (2µM). An XF Analyzer

(Seahorse Bioscience) was used to measure bioenergetics. Briefly, a XF96 extracellular flux assay cartridge (Seahorse Bioscience) was hydrated overnight at 37oC according to manufacturer’s instruction. XF assay medium supplemented with 25mM glucose, 10mM

o pyruvate and 0.3% fatty free acid BSA (pH 7.4) and incubated at 37 C in a non-CO2 incubator for 1 hour. Oligomycin (2µg/ml), carbonyl cyanide p- trifluoromethoxyphenylhydrazone (FCCP; 1µM), and Rotenone (2.5µM)/Antimycin A

(2.5µM)/2-DG (10mM) were sequentially injected and cellular oxygen consumption rate

(OCR) and extracellular acidification rate (ECAR) were quantified.

Hepatocyte culture

HepG2 cells were purchased from ATCC (#HB-8065) and tested negative for mycoplasma contamination. 1x105 HepG2 cells/condition were cultured in either the presence of adipocyte media conditioned with IFNb (250U/ml) and/or LPS (100ng/ml) or

DMEM/F:12 medium supplemented with IFNb (250U/ml), LPS (100ng/ml), IFNb/LPS for

48 hours. Culture supernatants were utilized for downstream processes.

Human Adipocyte Cytokine Quantification

Human primary adipocytes were cultured in the presence or absence of IFNβ (250U/ml) for 3 hours and thereafter stimulated with saline, LPS (100ng/ml), for 4 hours. IL-6

(Biologend) and IFNb (R&D systems) were quantified by ELISA as per manufacturer instructions.

Human Systemic Cytokine Quantification

Human IFNb, TNFa, IL-6, CXCL9, CCL3, and CXCL10 plasma concentrations from Met-

H and Met-C patients were determined by ELISA using MilliplexTM Multiplex kits

(MilliporeSigma) according to manufacturer’s protocol. Briefly, 25µL of plasma, plated in duplicate on a 96 well black plate, was incubated with 25µL of antibody coated beads.

Plates were washed and 25µL of secondary antibody was incubated, followed by 25µL of strept-avidin-RPE. 150µL of sheath fluid was added to plates that were washed and then

91 read using luminex technology on a Milliplex Analyzer (milliporeSigma). Data analysis performed by the Cincinnati Children’s Medical Center Research Flow Cytometry Core.

Statistical analysis

Statistical tests were utilized for all data sets with similar variance. Choice of test was dependent on number of groups and whether normal distribution exists. For all normally distributed data Student’s t-test was used for 2 groups whereas one-way ANOVA was utilized for 3 or more groups with Tukey’s post hoc test to determine differences between groups. All data presented as means +/- SEM. Analysis was performed via GraphPad

Prism Software’s. Determined sample sizes were based on preliminary data with respect to obesity modeling and type I IFN studies in myeloid cells including weight gain, immune cell infiltration, severity of obesity-associated sequelae, and interrogation of myeloid cell inflammatory vigor. No animals were excluded from the analyses and none of the studies were blinded.

100

101

102

103

104

105

106

107

108

109

110

Supplementary Fig. 1. Various TLR triggers are sufficient to promote adipocyte type I IFN axis. (a-c) Primary adipocytes and macrophages isolated from CD-fed WT mice. (a) IFNb protein; % change to macrophage. (b) mRNA expression of indicated type

I IFN axis genes, relative expression to macrophage. (c) IL-6 protein levels in stimulated macrophages and adipocytes under indicated conditions; % change to LPS stimulated macrophages. (d) IL-6 protein levels in stimulated adipocytes under indicated conditions;

% change to NS. (e) mRNA expression of indicated TLRs, relative expression to macrophage. (f) IL-6 protein levels in adipocytes treated with saline (NS), IFNb (200U/ml),

Poly(I:C) (50µM), Pam2cys (100ng/ml), or LPS (100ng/ml); % change to NS. (g-i) Primary adipocytes were isolated from WT mice and stimulated in the presence or absence of

Poly(I:C) (50µM). (g) IFNB1 mRNA expression. (h) IFNb protein. (i) Expression of indicated type I IFN axis genes, relative expression to NS. (a-c) Representative of 3 independent experiments, n = 3-4/condition. (d) A single experiment, n = 2/condition. (e)

Representative of 3 independent experiments, n = 3-4/condition. (f-i) Representative of 3 independent experiments, n = 3-5/condition. Data represents mean +/- SEM. (a-i)

Student’s t-test. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. M denotes macrophage and A denotes adipocyte.

Supplementary Fig. 2. Convergence of adipocytes and macrophage gene expression patterns. Primary adipocytes and macrophages isolated from CD fed WT mice were treated with saline (NS), IFNb (200 U/ml) or LPS (100 ng/ml) as indicated. (a- b) Venn diagram representation of number of differentially regulated genes in adipocytes and macrophages under NS. (c) Common pathways in adipocytes and macrophages

111 treated with IFNb+LPS. (d) Venn diagram representation of number of differentially regulated genes in adipocytes under indicated treatment. (a-d) A single experiment, n =

2/condition. A denotes adipocytes. M denotes macrophages.

Supplementary Fig. 3. Convergence of adipocyte and macrophage gene motif enrichments. (a-b) Complementary computational analyses of promoters of genes with elevated expression (determined by RNAseq) for TF binding site motif enrichment and

ChIP-seq and DNase-seq peak enrichment. (a) Promoter regions of genes expressed in

Adipocytes or Macrophages stimulated with saline (NS) or IFNb/LPS (200U/100ng/ml), were inspected for over-represented predicted TF binding sites (see Methods). Each data point in the scatterplot represents one TF binding site motif, with the X- and Y-axes indicating the negative log of the p-value describing the significance of the enrichment of the motif in the given gene promoter set. Motifs for particular TF classes are colored (see inset). Results are shown for promoters defined as (-1000,+500) relative to the TSS. NS, not stimulated. (b) Promoter regions of genes expressed in Adipocytes or Macrophages were inspected for over-represented ChIP-seq or DNase-seq peak datasets (see

Methods). Each data point in the scatterplot represents a single dataset (e.g., ChIP-seq for a particular TF or histone mark in a certain cell type). Particular types of datasets are colored (see inset). Results are shown for promoters defined as (-1000,+500) relative to the TSS. For both analyses, the X-axis shows adipocyte data sets and Y-axis displays macrophage data sets. (a-b) A single experiment, n = 2/condition. A denotes adipocytes.

M denotes macrophages.

112

Supplementary Fig. 4. IFNβ modifies glycolysis in adipocytes. (a-c) Adipocytes treated in the presence of absence of IFNβ (250U/ml). Adipocyte cellular bioenergetics

(oligomycin [10mM], FCCP [10mM], glucose [2mM]). (a) Oxygen consumption rate

(OCR). (b) Extracellular acidification rate (ECAR). (c) Quantified lactate in adipocyte supernatants. (d-e) Adipocyte cellular bioenergetics (oligomycin [10mM], FCCP [10mM], glucose [2mM]) treated in the presence or absence of IFNβ (250U/ml) and/or LPS

(100ng/ml). (d) OCR. (e) ECAR. (f-g) Adipocytes treated in the presence or absence of

Etomoxir (250μM). (f) OCR. (g) Quantified IL-6 protein in supernatants of (a-e)

Representative of 3 independent experiments, n = 2-3/condition. (f-g) Representative of

3 independent experiments, n = 3/condition. Data represents mean +/- SEM.

Supplementary Fig. 5. Obesity augments type I IFN signature across organs.

Expression of the indicated type I IFN axis genes in liver, eWAT and spleen of WT mice fed a chow diet (CD) or high-fat diet (HFD) for 22 weeks. Representative of 3 independent experiments, n = 6-7/condition. Data represents mean +/- SEM. Student’s t-test. **P <

0.01, ***P < 0.001, ****P < 0.0001.

Supplementary Fig. 6. IFNAR does not modify HFD-driven obesity and adiposity.

WT and IFNAR-/- mice were fed a HFD for 22 weeks. (a) Total body weight. (b) Energy

Expenditure. (c) Cumulative food intake between HFD-fed mice. (d) Systemic

Cholesterol. (e) Total body lean (%) and fat (%) mass. (f) Brown adipose tissue (BAT)

UCP-1 expression. (g) eWAT H&E staining. Representative of 3 independent

113 experiments, n = 6-7/condition. Data represents mean +/- SEM. (a) Area under curve. (c- f) Student’s t-test. ***P < 0.001, ****P < 0.0001.

Supplementary Fig. 7. IFNAR alters adipose tissue distribution. WT and IFNAR-/- mice were fed an HFD for 22 weeks. (a) eWAT weight. (b) inguinal WAT (iWAT) weight.

7/condition. Data represents mean +/- SEM. (a-c) Student’s t-test. *P < 0.05, **P < 0.01,

***P < 0.001, ****P < 0.0001.

Supplementary Fig. 8. IFNAR remodels WAT inflammation

WT and IFNAR-/- mice were fed an obesogenic diet for 22 weeks. eWAT indicated chemokines mRNA expression. Representative of 3 independent experiments, n = 6-

7/condition. Data represents mean +/- SEM. Student’s t-test. *P < 0.05, **P < 0.01.

Supplementary Fig. 9. IFNAR modulates obesity-associated NAFLD pathogenesis.

WT and IFNAR-/- mice were fed an obesogenic diet for 22 weeks. (a) H&E stained liver tissue. (b) Liver cholesterol and triglyceride levels. Representative of 3 independent experiments, n = 6-7/condition. Data represents mean +/- SEM, *P < 0.05. (b) Student’s t-test.

Supplementary Fig. 10. Non-hematopoietic and hematopoietic IFNAR expression contribute equally to obesity. Reciprocal bone marrow transfers (BMT) between WT and IFNAR-/- mice were performed, successful reconstitution was confirmed at d74 post-

114 transfer and mice were placed on a HFD for 18 weeks. (a) Reconstitution analysis by flow cytometry. (b) Body weight. (c) Liver weight. (d) eWAT, (e) iWAT and (f) pWAT weights.

A single experiment, n = 4-5/condition. Data represents mean +/- SEM. (b-f) Student’s t- test. *P < 0.05, **P < 0.01.

Supplementary Fig. 11. IFNb+LPS augments proinflammatory cytokine levels in adipocyte conditioned media. HepG2 cells were cultured in the presence of NS, IFNb

(200U/ml), LPS (100ng/ml) for 48 hours as indicated. (a) IL-6 protein; % change to NS.

(b) TNF protein; % change to NS. Data combined from 3 independent experiments, n =

2-3/condition. Data represents mean +/- SEM. (a-b) Student’s t-test. ***P < 0.001, ****P

< 0.0001.

Supplementary Fig. 12. Differentiated human adipocytes display a white adipocyte gene signature. Expression of indicated WAT signature genes, normalized arbitrary units to pAd. Representative patients, n = 4-5/condition. Data represents mean +/- SEM.

Student’s t-test. ****P < 0.0001. pAd denotes preadipocytes and Ad denotes adipocytes.

Supplementary Table 1. Obese WT and IFNAR-/- eWAT immune cell infiltration.

Supplementary Table 2. Obese WT and IFNAR-/- Liver immune cell infiltration.

Supplementary Table 3. Non-hematopoietic and hematopoietic eWAT immune cell infiltration.

115

Supplementary Table 4. Non-hematopoietic and hematopoietic liver immune cell infiltration.

Supplementary Table 5. Patient cohort demographics.

Supplementary Table 6. Met-H and Met-C systemic cytokine and chemokine levels.

116

Chapter 4. A BAFF/APRIL axis regulates obesogenic-diet driven weight gain

Calvin C. Chan1,2,7,8,#, Isaac T.W. Harley1,2,7,8,#, Paul T. Pfluger13,14,15,16, Aurelien Trompette1,2, Traci E. Stankiewicz1,2, Jessica L. Allen1,2,8, Maria E. Moreno-Fernandez1,2, Michelle S.M.A. Damen1,2, Matthew J. Flick1,3, Leah M. Flick1,2, Joan Sanchez- Gurmaches1,4,5 Rebekah Karns1,6, Michael Helmrath9,10, Thomas H. Inge11, Stuart P. Weisberg12, Sünje J. Pamp17,18, David A. Relman17,18,19, Randy J. Seeley20, Matthias H. Tschoep14,15,16, Christopher L. Karp1,2,7,8 and Senad Divanovic1,2,7,8,21*

1Department of Pediatrics, University of Cincinnati College of Medicine, Cincinnati, OH 45220. Divisions of 2Immunobiology, 3Experimental Hematology, 4Endocrinology, 5Developmental Biology, and 6Gastroenterology, Hepatology and Nutrition, Cincinnati Children’s Hospital Medical Center Cincinnati, OH 45229, USA. 7Medical Scientist Training Program and 8Immunology Graduate Program Cincinnati Children’s Hospital Medical Center and the University of Cincinnati College of Medicine, Cincinnati, OH 45220, USA. 9Pediatric General and Thoracic Surgery, Cincinnati Children’s Hospital Medical Center, Cincinnati, OH 45229. 10Stem Cell & Organoid Medicine, Cincinnati Children’s Hospital Medical Center, Cincinnati, Ohio, USA 45229; 11Department of Surgery, Children’s Hospital Colorado, Aurora CO 80045, USA; 12Columbia University Medical Center, New York, NY 10032, USA; 13Research Unit NeuroBiology of Diabetes and 14Institute for Diabetes and Obesity, Helmholtz Center Munich, 85764 Neuherberg, Germany; 15German Center for Diabetes Research (DZD), 85764 Neuherberg, Germany; 16Division of Metabolic Diseases, Technische Universität München, 80333 Munich, Germany; Departments of 17Microbiology and Immunology, and 18Medicine, Stanford University, Stanford, CA 94305, USA; 19Veterans Affairs Palo Alto Health Care System, Palo Alto, CA 94304, USA; 20Departments of Surgery, Internal Medicine and Nutritional Sciences, University of Michigan, Ann Arbor, MI 48109, USA; 21Center for Inflammation and Tolerance, Cincinnati Children’s Hospital Medical Center, Cincinnati, Ohio, USA 45229.

#Denotes equal contribution *Correspondence: [email protected]

Current Addresses:

Isaac T.W. Harley: Division of Rheumatology, Department of Internal Medicine and Department of Immunology & Microbiology, University of Colorado Denver, Aurora, CO 80045, USA.

Aurelien Trompette: University of Lausanne, Service de Pneumologie, CHUV, CLED 02.206, Chemin des Boveresses 155, 1066 Epalinges, Switzerland.

Jessica L. Allen: Charlotte, NC 28270.

117

Matthew J. Flick: The University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, USA.

Leah M. Flick: Meridian Bioscience, Inc., Cincinnati, OH 45244, USA.

Sünje J. Pamp: National Food Institute, Technical University of Denmark, Kgs. Lyngby, Denmark.

Christopher L. Karp: Global Health Discovery & Translational Sciences, and Maternal Neonatal and Child Health—Discovery & Tools, the Bill & Melinda Gates Foundation, Seattle, WA 98102, USA.

118

ABSTRACT

Despite growing appreciation of complex, physiologically important interactions between the immune system and adipose tissue (AT), the impact of immune mediators on weight homeostasis via modulation of AT function remains underdefined. Mechanistic interrogation of resistance to diet-induced obesity (DIO) in mice lacking a negative regulator of Toll-like receptor (TLR) signaling serendipitously uncovered a prominent role for B cell activating factor (BAFF) in regulating weight gain and AT lipid handling. Notably, overexpression of BAFF in multiple transgenic mouse lines was associated with protection from weight gain, approximating a log-linear dose response relation to serum

BAFF concentrations. Global gene expression analysis of BAFF-stimulated adipocytes revealed upregulation of pathways regulating lipid metabolism. Congruently, white adipose tissue (WAT) from BAFF-overexpressing mice exhibited enhanced lipolysis, with

BAFF being sufficient to induce white adipocyte lipolysis in vitro. Brown adipose tissue

(BAT) from BAFF-overexpressing mice also exhibited increased UCP-1 expression, with

BAFF being sufficient to augment brown adipocyte respiration in vitro and energy expenditure in vivo. APRIL, a close homolog of BAFF, similarly modulated both WAT and

BAT lipid handling. While genetic deletion of BAFF or APRIL alone did not augment weight gain, genetic deletion of both BAFF and APRIL exacerbated DIO in mice. Lastly, these effects of BAFF and APRIL were conserved in human adipocytes and higher systemic BAFF/APRIL levels were correlated with greater decrease in BMI after bariatric surgery. Thus, the BAFF/APRIL axis is a novel, multifaceted immune-driven regulator of

AT function and body weight control.

119

INTRODUCTION

Obesity is a major and unabated public health problem1-3. Obesity stems from an imbalance between energy intake (e.g., food intake, microbiome, nutrient absorption), energy expenditure (e.g., basal metabolism, thermogenesis) and energy turnover

(lipolysis, lipogenesis)4,5. Brown adipose tissue (BAT) and white adipose tissue (WAT) are key contributors to controlling body weight6. BAT, a highly energy consuming tissue rich with mitochondria, is capable of removing lipids from circulation to activate adaptive thermogenesis7. WAT can assimilate calories through storage of lipids into adipocytes.

WAT lipolysis facilitates the breakdown of lipids into functional consumable units including glycerol8. Persistent imbalances of caloric intake and energy expenditure leads to the pathological expansion of adipose tissue (AT) that contributes to the propagation of obesity-associated low-grade inflammation. This maladaptive obesity-associated inflammation underlies the pathogenesis of many obesity-associated sequelae, including type 2 diabetes mellitus, atherosclerosis, and non-alcoholic fatty liver disease

(NAFLD)9,10. In light of the clinical and public health challenges resulting from the limitations of current methods for controlling weight gain, a better understanding of the pathogenesis is required. Leveraging this understanding may lead to new innovative approaches targeting body weight11,12.

Obesity is associated with an increase in circulating TLR ligands (e.g., lipopolysaccharide [LPS]13). However, the contribution of TLRs, including TLR4, to obesity pathogenesis remains controversial with both beneficial14 and detrimental15-18 roles reported. TLR4 signaling in myeloid cells is negatively regulated by radioprotective

105kDa protein (RP105; CD180)19, a molecule originally thought to be B cell-specific and

120 to facilitate B cell proliferation19,20, in response to LPS. Investigation of these apparently dichotomous cell type specific effects of RP105 on TLR4 signaling identified that RP105- mediated regulation of B cell proliferation was not due to B cell intrinsic RP105 expression. Rather, chronic upregulation of B cell activating factor (BAFF) expression was revealed as the mechanism underlying suppressed B cell proliferation in response to LPS in RP105-deficient mice20. Surprisingly, despite its known actions as a negative regulator of TLR-driven inflammation, RP105-deficiency dampens AT inflammation and protects mice from obesity-associated metabolic derangements21. However, the mechanisms underlying protection in RP105 deficient mice have not been defined.

Both TLR4 and RP105 modulate the production of inflammatory mediators including the tumor necrosis factor (TNF) superfamily members (e.g., TNF, BAFF)20,22.

TNF can induce white adipocyte lipolysis23,24 and may modulate brown adipocyte thermogenesis25-28. Although recent studies highlight the ability of immune-modulators to regulate AT function29-32, the link between immune mediators, specifically TNF family members, weight gain, and BAT/WAT function remains poorly understood. BAFF and its close homolog A proliferation-inducing ligand (APRIL) are regulators of B cell maturation and survival33. BAFF and APRIL are produced by diverse myeloid, lymphoid, and non- hematopoietic cell types34, including adipocytes35, astrocytes36, and gut epithelial cells37.

Human and mouse adipocytes express all known BAFF and APRIL receptors including

BAFF receptor (BAFF-R), transmembrane activator and CAML interactor (TACI) and B- cell maturation protein (BCMA). While BAFF can bind all three receptors, APRIL only binds to TACI and BCMA33. Recent reports indicate that BAFF is a regulator of atherosclerosis, and BAFF-R and TACI regulate obesity-associated metabolic

121 sequelae38-41. Whether a BAFF/APRIL axis contributes to the regulation of weight gain in

DIO models has remained undefined.

Our exploration of mechanisms regulating protection from DIO in RP105-deficient mice led to the serendipitous discovery of a critical role of BAFF and APRIL in the regulation of weight gain. Here, our collective data highlight a novel pathway of immune- mediated regulation of AT/adipocyte lipid handling and may represent novel targets for the treatment of obesity.

122

RESULTS

High fat diet (HFD)-fed RP105-/- mice were markedly protected from DIO and obesity- associated metabolic sequelae as compared to WT counterparts (Fig. 1a; and data not shown [DNS]), in agreement with a previous report21. Exploration of direct underlying mechanisms driving unexpected resistance to DIO in RP105-/- mice revealed that the effect was not associated with differential energy intake/metabolism, fat absorption, leptin sensitivity, or intestinal microbiome composition (Supplemental Fig. 1). Employment of bone marrow transfer studies suggested that lack of RP105 expression in the non- hematopoietic compartment was critical for protective effects of RP105 genetic ablation

(Supplemental Fig. 2a-b). Generation and utilization of effective, conditional RP105 deletion (RP105flox/flox mice) in total immune cells (Vav1cre), B cells (CD19cre), skeletal muscle (MLCcre), or central/peripheral nervous system (Nestincre) failed to fully reveal a primary RP105-expressing cellular locus responsible for observed differential weight gain

(Supplemental Fig. 2c-g). Collectively, these findings suggested the possibility of an indirect mechanism underlying protection from DIO in RP105-/- mice.

Given the functional relevance of BAFF in B cell responsiveness in RP105-/- mice20, we intuitively hypothesized that BAFF, which can be induced by multiple proinflammatory means (e.g., TLR signaling34), may also affect DIO in RP105-/- mice. Deletion of BAFF in

RP105-/- mice (RP105-/-/BAFF-/-) was sufficient to reverse the protection from HFD-driven weight gain, fasting glucose, GTT, and fat pad distribution as compared to WT controls

(Fig. 1b-d; Supplemental Fig. 3). These observations indicated that BAFF may be a regulator of weight homeostasis observed in RP105-/- mice.

123

To define whether the impact of BAFF as a regulator of weight gain extends beyond the RP105 system we utilized model systems that enhance expression of BAFF.

While RP105-/- mice marginally overexpress BAFF (~20% increase), B cell deficient mice

(µMT) exhibit 2 log-fold secondary increase in BAFF, and BAFF-Tg mice consists of a 3 log-fold increase in BAFF (Fig. 1e). Enhanced systemic BAFF levels were associated with increased resistance to HFD-driven weight gain in an approximate log-linear dose fashion (Fig. 1e-f). HFD-fed BAFF-Tg (Fig. 1g-j) and µMT (Supplemental Fig. 4) mice were protected from weight-driven glucose dysmetabolism (e.g., fasting glucose and

GTT; Fig. 1g-h) and NAFLD progression (hepatic triglyceride levels [Fig. 1i] and alanine transaminase [ALT; Fig. 1j]). Overall these findings suggest that increased circulating

BAFF may regulate DIO-driven weight gain.

Obesity pathogenesis/weight gain is regulated, in part, by adipocyte lipid handling.

Inflammatory mediators (e.g., TNF) can regulate lipid processing23,30. BAFF, a TNF superfamily member, and all three known BAFF receptors (e.g., BAFF-R, TACI, BCMA) are expressed by both mouse and human adipocytes35. Thus, we examined the ability of

BAFF to modulate adipocyte function. Utilization of an unbiased RNA-seq analysis revealed that recombinant BAFF (rBAFF), compared to saline treated counterparts, upregulated lipid handling pathways, including metabolism of lipids and lipoproteins, regulation of lipid metabolic process, neutral lipid metabolic process and cellular lipid metabolic process (Fig. 2a) in primary adipocytes. In line with these observations, WAT from RP105-/- and BAFF-Tg mice had significantly increased mRNA expression of adipose triglyceride lipase (PNPLA2) and hormone sensitive lipase (LIPE), both critical regulators of lipolysis (Fig. 2b). Enhanced expression of PNPLA2 and LIPE correlated

124 with increases in systemic BAFF levels (Fig. 1e). We next tested the direct impact of

BAFF on white adipocyte lipolysis. rBAFF was sufficient to augment the presence of free glycerol in adipocyte supernatants (Fig. 2c) and to enhance the mRNA expression of both

PNPLA2 and LIPE in vitro and in vivo (Fig. 2d-e). Notably, genetic deletion of BAFF-R,

TACI, or BCMA abrogated rBAFF-driven white adipocyte lipolysis in vitro (Fig. 2f). rBAFF- driven lipolysis was specific to white, but not brown, adipocytes (Supplemental Fig. 5a).

Combined, these findings indicate that BAFF is capable of modulating WAT/white adipocyte lipolysis.

BAT/brown adipocyte thermogenesis is a key regulator of lipid utilization and weight control. Thus, we also examined whether BAFF could impact BAT thermogenic capacity. Thermogenin (UCP-1) promotes adaptive thermogenesis in BAT42. RP105-/-

BAT and BAFF-Tg BAT exhibited increased UCP-1 expression compared to WT controls

(Fig. 3a). BAT mitochondria from these mice exhibited significantly enhanced respiration as compared to WT controls (Fig. 3b). Congruently, rBAFF was sufficient to amplify UCP-

1 expression (Fig. 3c), enhance respiration in primary brown adipocytes in vitro (Fig. 3d) and raise the threshold of norepinephrine-driven thermogenic activity (Fig. 3e) in WT mice in vivo. These effects were specific to brown adipocytes as rBAFF did not enhance white adipocyte UCP-1 expression (Supplemental Fig. 5b). We additionally assessed the sufficiency of exogenous BAFF to modify BAT and thermogenic parameters in vivo. In contrast to pre-treatment (Supplemental Fig. 6a), short-term exogenous rBAFF in CD- fed mice transiently amplified energy expenditure (EE), specifically in the dark cycles

(Supplemental Fig. 6b-c), but did not alter the respiratory exchange ratio (RER) or body weight (Supplemental Fig. 6d-e). Congruently, rBAFF administration to HFD-fed mice

125 enhanced total EE (Fig. 3f-g). Collectively, these findings indicate that BAFF, in parallel to its effects on WAT/white adipocytes, is capable of modulating BAT/brown adipocyte thermogenesis.

Examination of HFD-fed BAFF-/- mice revealed no alteration in weight gain (Fig.

4a), slightly increased body weight (Supplemental Fig. 7a), similar fasting glucose

(Supplemental Fig. 7b) and protection from hepatocellular injury (Supplemental Fig.

7c), consistent with previous reports43,44. This pointed towards the possibility that other

BAFF-like molecules may similarly modulate weight gain to potentially compensate for such effects in the absence of BAFF. Notably, BAFF-/- mice exhibited enhanced circulating levels of APRIL (Fig. 4b), a close homolog of BAFF33. Congruent with similarities of BAFF and APRIL function on immune cells, an unbiased RNA-seq approach revealed significant gene overlap (52% of BAFF-induced genes) between BAFF or APRIL treated primary white adipocytes (769 genes; Fig. 4c). Pathway analyses revealed that

APRIL stimulation of white adipocytes also upregulated pathways associated with lipolysis including regulation of lipid metabolism, fatty acid metabolism, PPAR signaling, and the glucocorticoid receptor regulatory network (Fig. 4d). Like BAFF, APRIL was sufficient to augment white adipocyte free glycerol release and induce expression of LIPE and PNPLA2 in vitro and in vivo (Fig. 4e-f). Notably, lack of TACI or BCMA, but not BAFF-

R, prohibited APRIL-driven lipolysis in vitro (Fig. 4g)— as would be expected from the pattern of receptor usage by APRIL. Examination of APRIL’s effects on brown adipocytes demonstrated sufficiency to augment UCP-1 expression and enhance mitochondrial respiration (Fig. 4h-i). However, exogenous rAPRIL administration did not amplify energy expenditure in lean or obese mice (Supplemental Fig. 8). Taken together, these findings

126 indicate that APRIL, similar to BAFF, is capable of modulating both white and brown adipocyte lipid handling.

Although close homologs with some overlapping function, deeper examination of our unbiased RNA-seq analyses displayed a large number (2535) of genes modulated by

APRIL, as compared to BAFF in primary adipocytes (Fig. 4c). APRIL specifically downregulated inflammatory pathways including IL-6 mediated signaling, Toll-like receptor signaling pathway, and inflammation mediated by chemokine and cytokine signaling pathway (Supplemental Fig. 9a). Screening of obesity-associated proinflammatory cytokines and chemokines in adipocytes demonstrated that BAFF, but not

APRIL, upregulated the mRNA expression of multiple pro-inflammatory mediators, including TNF, CXCL1, CCL2, CCL3 (Supplemental Fig. 9b-e). Thus, these findings suggest that despite similar effects to modify adipocyte lipid handling, APRIL, unlike

BAFF, may not induce an inflammatory response in adipocytes.

As APRIL modifies adipocyte lipid handling, we next tested the effect of genetic deletion of APRIL on DIO. Surprisingly, APRIL-/- mice likewise exhibited augmented levels of systemic BAFF – further suggestive that deletion of BAFF or APRIL alone drives compensatory increases of the reciprocal family member (BAFF-/- Fig. 4b; APRIL-/-; Fig.

5a). Consistent with the overarching observation of increased BAFF association with weight control (Fig. 1e-f), APRIL-/- mice were resistant to DIO (Fig. 5b), despite similar food intake (Fig. 5c), as compared to WT controls. This protection was associated with higher energy expenditure and decreased metabolic sequelae, including glucose dysmetabolism and hepatocellular damage (Fig. 5d-h). Notably, combined deletion of

BAFF and APRIL in mice thoroughly unveiled the contribution of this axis in DIO by

127 amplification of weight gain above the levels of BAFF-/-, APRIL-/-, and WT mice (Fig. 5i).

Combined, these data indicate a prominent role for the BAFF/APRIL axis in the development of obesity (Fig. 1f, 5i).

To determine if BAFF and APRIL’s effects were conserved in humans, primary human white adipocytes were isolated from WAT of severely obese individuals undergoing surgical treatment (Supplemental Table 1). Consistent with our findings in mice, rBAFF and rAPRIL treatment resulted in enhanced human adipocyte free glycerol release (Fig. 6a) and induced the mRNA expression of PNPLA2 and LIPE (Fig. 6b-c). At baseline, persons with severe obesity had significantly lower levels of circulating BAFF but increased levels of APRIL as compared to lean controls (Fig. 6d-e). Similarly, BAFF, but not APRIL, was negatively correlated with increased BMI (Fig. 6f-g). Notably, one- year after bariatric surgery and after substantial weight loss (Supplemental Table 1),

BAFF and APRIL levels were positively correlated with a greater change in BMI (Fig. 6h- i). In unison, these findings suggest that the BAFF/APRIL axis exerts conserved effects on human adipocytes and may be a potential surrogate marker for the degree of weight loss after bariatric surgery.

128

DISCUSSION

The traditional perception of inflammation in the context of obesity has focused on the detrimental impact of inflammatory mediators on the pathogenesis of secondary metabolic derangements associated with obesity. While inflammation undoubtedly exacerbates obesity-associated sequelae, the direct relevance to the regulation of weight gain remains underdefined. Our collective findings presented here highlight the beneficial impact of the BAFF/APRIL axis to control weight gain by altering metabolic regulation of

AT and energy homeostasis (Fig. 7). In fact, these findings add to the growing number of reports invoking a potentially advantageous contribution of inflammation to obesity development14,30. Body weight control and pathogenesis of obesity-associated sequelae are interlinked processes. RNA-seq analysis of BAFF or APRIL treated adipocytes revealed alterations to pathways regulating lipid metabolic processes (Fig. 2a, 4d). Mice with increased circulating BAFF exhibited enhanced WAT expression of lipolytic mediators (Fig. 2b), but are protected from metabolic sequalae associated with increased circulating lipids45,46. This may hint at the existence of evolutionary mechanisms to handle circulating lipids and protect from metabolic derangements – mechanisms possibly involving the BAFF/APRIL axis.

BAFF and APRIL are traditionally perceived as regulators of B cell maturation and survival. B cells infiltrate AT in the context of obesity and have been suggested to contribute to the downstream metabolic derangements of obesity47. Our data indicate that multiple lines of mice with increased systemic BAFF, including RP105-/- or BAFF-Tg

(augmented B cell numbers) and µMT or APRIL-/- mice (devoid or decreased number of mature B cells) are similarly protected from DIO and the subsequent downstream

129 metabolic sequelae (Fig. 1 and Fig. 5). The consistent unifying factor among these different transgenic lines are elevated systemic BAFF levels, which suggests that our observed effects of BAFF may be independent of B cells. In contrast to our findings, existing reports indicate that µMT-/- mice gain similar weight47 to WT counterparts.

Nonetheless, reported protection from obesity-associated glucose dysmetabolism47 are congruent with our findings. As B cells play critical roles in the establishment of the gut microbiota (e.g., IgA regulation48), it is plausible that lack of B cells amplifies differential colonization within µMT-/- mice harbored at different facilities.

Despite similar capacities of BAFF and APRIL to modify white and brown adipocyte lipid handling, we demonstrate that BAFF-/- or APRIL-/- mice do not exhibit increased obesogenic diet-driven weight gain (Fig. 4a, 5b). However, BAFF-/- mice exhibit approximately 1.5-fold (48%) increase in APRIL, while APRIL-/- mice similarly exhibit approximately 2-fold (59%) increase in systemic BAFF (Fig. 4b, 5a) – likely stemming from the activation of compensatory mechanisms. Our findings that lean individuals have higher levels of BAFF than obese individuals, but lower APRIL levels, suggest possible reciprocal regulation between BAFF and APRIL. Notably, some inflammatory mediators within the same family (e.g., IL-17a and IL-17f)49 can regulate each other’s expression.

Thus, examination of the interlink between BAFF and APRIL expression is warranted.

These collective data pertaining to the BAFF/APRIL axis, however, hint that the presence of increased BAFF or APRIL can drive resistance to DIO. Consistent with this idea, lack of both BAFF and APRIL exacerbated DIO-driven weight gain (~50% increase Fig. 5i) suggesting that the BAFF/APRIL system likely works in concert to regulate weight gain.

130

Recent reports suggest that BAFF and APRIL’s receptors, including BAFF-R and

TACI, regulate obesity pathogenesis38-41. While BAFF-R-/- or TACI-/- mice gain significantly more weight on a HFD38-41, these mice are protected from the downstream glucose dysmetabolism. These observations are in line with our findings that the

BAFF/APRIL axis is tied to body weight control. Further, HFD-fed BAFF-R-/- exhibit enhanced steatosis41. As decreased BAT activity is linked with development of steatosis50-52, it is plausible that augmented steatosis in BAFF-R-/- mice consists of dampened BAT activity or even direct effect of BAFF on hepatocytes via BAFF-R signaling. In contrast, the impact of BCMA on weight gain remains unreported and would provide more clues to the relevance of this system in different aspects of obesity development.

Our findings indicate that BAFF-R impacts BAFF-driven white adipocyte lipolysis, while TACI and BCMA modulates BAFF and APRIL-induced lipolysis – findings in line with the selectivity in BAFF and APRIL binding to these receptors (Fig. 2e, 4f). As a single receptor deletion (BAFF-R-/-, TACI-/-, BCMA-/-) abrogates BAFF or APRIL-driven lipolysis, this suggests the possibility that engagement of all three receptors on white adipocytes is necessary to provide sufficient signal strength for activation of lipolysis. Hence, closer examination of the critical BAFF/APRIL receptor(s) involved with its capacity to control body weight would be of significant future interest.

BAFF/APRIL effects on lipolysis are distinctive in white adipocytes, while induction of UCP-1 is specific to brown adipocytes (Supplementary Fig. 5). This layered complexity of the BAFF/APRIL system (2 molecules and 3 receptors) invokes the possibility of a highly regulated mechanism(s) to impact function across different

131 tissues/cell types (e.g., WAT, BAT, liver). Additionally, although our findings suggest responsiveness by WAT and BAT to BAFF and APRIL, the relevant source of BAFF and/or APRIL in the context of obesogenic-diet remains unknown. Both hematopoietic and non-hematopoietic cells, including adipocytes, can produce BAFF and APRIL34,35,53.

Complicating matters, BAFF and APRIL exist as oligomers, homo/hetero multimers and their conformation regulates the capacity of BAFF and APRIL to activate their three receptors.54 Future exploration of the conformation and relevant sources of BAFF and

APRIL in the context of obesity and control of body weight should be investigated.

BAFF and APRIL have the capacity to modify DIO in mice (Fig. 1f and 5i). In addition, the effects of both BAFF and APRIL are conserved in primary human adipocytes. Novel findings of a direct correlation between higher BAFF and APRIL levels with greater changes in BMI after bariatric surgery indicate that the BAFF and/or APRIL axis may play a role in human weight regulation after bariatric surgery and could represent a prognostic marker for weight loss outcomes. As the BAFF and/or APRIL axis protects mice from DIO, BAFF and/or APRIL may also have therapeutic potential in humans. In this regard, even a minimal increase in BAFF (~20% increase in RP105-/- mice) would be protective of DIO and would likely not cause the deleterious effects observed with log- increases of BAFF55,56 (e.g., autoimmunity, B cell-driven pathology).

In contrast to BAFF, APRIL does not appear to induce a proinflammatory program in adipocytes (Supplemental Fig. 9) while maintaining its capacity to modulate white and brown adipocyte lipid handling and play a role in protection from DIO (Fig. 4-6). Thus, an approach harnessing the BAFF/APRIL axis may need careful monitoring of the levels of

BAFF/APRIL in order to provide a maximal therapeutic index. Further definition of the

132 segregation between BAFF and APRIL in adipocytes, including the critical receptors involved and downstream signaling mechanisms is needed.

Our findings and existing literature hint at a potentially dichotomous relationship presented by inflammation in obesity: (a) inflammatory cytokines that can regulate development of weight gain and (b) established obesity-driven inflammatory cytokines that can amplify downstream metabolic derangements. Thus, a balance between these two functions of the immune system likely needs to be tightly controlled and regulated.

We predict that cytokines, including BAFF and APRIL, can act to mobilize energy from

AT in times requiring increased energy expenditure and lipid handling (e.g., obesity development, infections57-61). However, over activation of these inflammatory pathways, including the BAFF/APRIL axis, can promote type 2 diabetes and autoimmune disorders

(e.g., Systemic Lupus Erythematosus, Sjogren’s Syndrome)34. Notably, involuntary weight loss62 is a known symptom accompanying some autoimmune diseases.

In summary (Fig. 7), our data demonstrate the impact of the BAFF and APRIL axis as beneficial mediators of white and brown adipocyte lipid handling and body weight control. These observations may highlight a prospective and appealing pathway open to the development of new clinical approaches for body weight modulation.

133

REFERENCES

1. Finucane, M.M., et al. National, regional, and global trends in body-mass index since 1980: systematic analysis of health examination surveys and epidemiological studies with 960 country-years and 9.1 million participants. Lancet 377, 557-567 (2011). 2. Flegal, K.M., Carroll, M.D., Ogden, C.L. & Curtin, L.R. Prevalence and trends in obesity among US adults, 1999-2008. JAMA 303, 235-241 (2010). 3. Global, B.M.I.M.C. Body-mass index and all-cause mortality: individual-participant- data meta-analysis of 239 prospective studies in four continents. Lancet (2016). 4. Tseng, Y.H., Cypess, A.M. & Kahn, C.R. Cellular bioenergetics as a target for obesity therapy. Nature reviews. Drug discovery 9, 465-482 (2010). 5. Flier, J.S. Obesity wars: molecular progress confronts an expanding epidemic. Cell 116, 337-350 (2004). 6. Seale, P. & Lazar, M.A. Brown fat in humans: turning up the heat on obesity. Diabetes 58, 1482-1484 (2009). 7. Bartelt, A., et al. Brown adipose tissue activity controls triglyceride clearance. Nat Med 17, 200-205 (2011). 8. Saponaro, C., Gaggini, M., Carli, F. & Gastaldelli, A. The Subtle Balance between Lipolysis and Lipogenesis: A Critical Point in Metabolic Homeostasis. Nutrients 7, 9453-9474 (2015). 9. Hotamisligil, G.S. Inflammation, metaflammation and immunometabolic disorders. Nature 542, 177-185 (2017). 10. Senn, J.J., Klover, P.J., Nowak, I.A. & Mooney, R.A. Interleukin-6 induces cellular insulin resistance in hepatocytes. Diabetes 51, 3391-3399 (2002). 11. Wing, R.R. & Phelan, S. Long-term weight loss maintenance. Am J Clin Nutr 82, 222S-225S (2005). 12. Melnikova, I. & Wages, D. Anti-obesity therapies. Nature reviews. Drug discovery 5, 369-370 (2006). 13. Cani, P.D., et al. Metabolic endotoxemia initiates obesity and insulin resistance. Diabetes 56, 1761-1772 (2007). 14. Tao, C., et al. Short-Term Versus Long-Term Effects of Adipocyte Toll-Like Receptor 4 Activation on Insulin Resistance in Male Mice. Endocrinology 158, 1260-1270 (2017). 15. Vila, I.K., et al. Immune cell Toll-like receptor 4 mediates the development of obesity- and endotoxemia-associated adipose tissue fibrosis. Cell Rep 7, 1116- 1129 (2014). 16. Jia, L., et al. Hepatocyte Toll-like receptor 4 regulates obesity-induced inflammation and insulin resistance. Nat Commun 5, 3878 (2014). 17. Vijay-Kumar, M., et al. Metabolic syndrome and altered gut microbiota in mice lacking Toll-like receptor 5. Science 328, 228-231 (2010). 18. Wu, L.H., et al. Loss of toll-like receptor 3 function improves glucose tolerance and reduces liver steatosis in obese mice. Metabolism 61, 1633-1645 (2012). 19. Divanovic, S., et al. Negative regulation of Toll-like receptor 4 signaling by the Toll- like receptor homolog RP105. Nat Immunol 6, 571-578 (2005).

134

20. Allen, J.L., et al. Cutting edge: regulation of TLR4-driven B cell proliferation by RP105 is not B cell autonomous. J Immunol 188, 2065-2069 (2012). 21. Watanabe, Y., et al. The radioprotective 105/MD-1 complex contributes to diet- induced obesity and adipose tissue inflammation. Diabetes 61, 1199-1209 (2012). 22. Chu, V.T., Enghard, P., Riemekasten, G. & Berek, C. In vitro and in vivo activation induces BAFF and APRIL expression in B cells. J Immunol 179, 5947-5957 (2007). 23. Zhang, H.H., Halbleib, M., Ahmad, F., Manganiello, V.C. & Greenberg, A.S. Tumor necrosis factor-alpha stimulates lipolysis in differentiated human adipocytes through activation of extracellular signal-related kinase and elevation of intracellular cAMP. Diabetes 51, 2929-2935 (2002). 24. Souza, S.C., et al. TNF-alpha induction of lipolysis is mediated through activation of the extracellular signal related kinase pathway in 3T3-L1 adipocytes. J Cell Biochem 89, 1077-1086 (2003). 25. Masaki, T., et al. Tumor necrosis factor-alpha regulates in vivo expression of the rat UCP family differentially. Biochim Biophys Acta 1436, 585-592 (1999). 26. Coombes, R.C., Rothwell, N.J., Shah, P. & Stock, M.J. Changes in thermogenesis and brown fat activity in response to tumour necrosis factor in the rat. Biosci Rep 7, 791-799 (1987). 27. Nisoli, E., et al. Tumor necrosis factor alpha mediates apoptosis of brown adipocytes and defective brown adipocyte function in obesity. Proc Natl Acad Sci U S A 97, 8033-8038 (2000). 28. Valladares, A., Roncero, C., Benito, M. & Porras, A. TNF-alpha inhibits UCP-1 expression in brown adipocytes via ERKs. Opposite effect of p38MAPK. FEBS Lett 493, 6-11 (2001). 29. Green, A., Rumberger, J.M., Stuart, C.A. & Ruhoff, M.S. Stimulation of lipolysis by tumor necrosis factor-alpha in 3T3-L1 adipocytes is glucose dependent: implications for long-term regulation of lipolysis. Diabetes 53, 74-81 (2004). 30. Wernstedt Asterholm, I., et al. Adipocyte inflammation is essential for healthy adipose tissue expansion and remodeling. Cell Metab 20, 103-118 (2014). 31. Eguchi, J., et al. Transcriptional control of adipose lipid handling by IRF4. Cell Metab 13, 249-259 (2011). 32. Eguchi, J., et al. Interferon regulatory factors are transcriptional regulators of adipogenesis. Cell Metab 7, 86-94 (2008). 33. Mackay, F., Schneider, P., Rennert, P. & Browning, J. BAFF AND APRIL: a tutorial on B cell survival. Annu Rev Immunol 21, 231-264 (2003). 34. Mackay, F. & Schneider, P. Cracking the BAFF code. Nat Rev Immunol 9, 491- 502 (2009). 35. Alexaki, V.I., et al. Adipocytes as immune cells: differential expression of TWEAK, BAFF, and APRIL and their receptors (Fn14, BAFF-R, TACI, and BCMA) at different stages of normal and pathological adipose tissue development. J Immunol 183, 5948-5956 (2009). 36. Krumbholz, M., et al. BAFF is produced by astrocytes and up-regulated in multiple sclerosis lesions and primary central nervous system lymphoma. J Exp Med 201, 195-200 (2005).

135

37. Wang, Y., et al. An LGG-derived protein promotes IgA production through upregulation of APRIL expression in intestinal epithelial cells. Mucosal Immunol 10, 373-384 (2017). 38. Jackson, S.W., et al. Cutting Edge: BAFF Overexpression Reduces Atherosclerosis via TACI-Dependent B Cell Activation. J Immunol 197, 4529-4534 (2016). 39. Liu, L., et al. TACI-Deficient Macrophages Protect Mice Against Metaflammation and Obesity-Induced Dysregulation of Glucose Homeostasis. Diabetes 67, 1589- 1603 (2018). 40. Tsiantoulas, D., et al. BAFF Neutralization Aggravates Atherosclerosis. Circulation (2018). 41. Kawasaki, K., et al. Blockade of B-cell-activating factor signaling enhances hepatic steatosis induced by a high-fat diet and improves insulin sensitivity. Lab Invest 93, 311-321 (2013). 42. Yang, X. & Ruan, H.B. Neuronal Control of Adaptive Thermogenesis. Front Endocrinol (Lausanne) 6, 149 (2015). 43. Kim, D.H. & Do, M.S. BAFF knockout improves systemic inflammation via regulating adipose tissue distribution in high-fat diet-induced obesity. Exp Mol Med 47, e129 (2015). 44. Nakamura, Y., et al. Depletion of B cell-activating factor attenuates hepatic fat accumulation in a murine model of nonalcoholic fatty liver disease. Sci Rep 9, 977 (2019). 45. Boden, G. Free fatty acids-the link between obesity and insulin resistance. Endocr Pract 7, 44-51 (2001). 46. Kahn, B.B. & Flier, J.S. Obesity and insulin resistance. J Clin Invest 106, 473-481 (2000). 47. DeFuria, J., et al. B cells promote inflammation in obesity and type 2 diabetes through regulation of T-cell function and an inflammatory cytokine profile. Proc Natl Acad Sci U S A 110, 5133-5138 (2013). 48. Moon, C., et al. Vertically transmitted faecal IgA levels determine extra- chromosomal phenotypic variation. Nature 521, 90-93 (2015). 49. von Vietinghoff, S. & Ley, K. IL-17A controls IL-17F production and maintains blood neutrophil counts in mice. J Immunol 183, 865-873 (2009). 50. Mottillo, E.P., et al. Lack of Adipocyte AMPK Exacerbates Insulin Resistance and Hepatic Steatosis through Brown and Beige Adipose Tissue Function. Cell Metab 24, 118-129 (2016). 51. Poekes, L., et al. Defective adaptive thermogenesis contributes to metabolic syndrome and liver steatosis in obese mice. Clin Sci (Lond) 131, 285-296 (2017). 52. Shen, H., Jiang, L., Lin, J.D., Omary, M.B. & Rui, L. Brown fat activation mitigates alcohol-induced liver steatosis and injury in mice. J Clin Invest 130(2019). 53. Kim, Y.H., Choi, B.H., Cheon, H.G. & Do, M.S. B cell activation factor (BAFF) is a novel adipokine that links obesity and inflammation. Exp Mol Med 41, 208-216 (2009). 54. Bossen, C., et al. TACI, unlike BAFF-R, is solely activated by oligomeric BAFF and APRIL to support survival of activated B cells and plasmablasts. Blood 111, 1004- 1012 (2008).

136

55. Maglione, P.J., et al. BAFF-driven B cell hyperplasia underlies lung disease in common variable immunodeficiency. JCI Insight 4(2019). 56. Moisini, I. & Davidson, A. BAFF: a local and systemic target in autoimmune diseases. Clin Exp Immunol 158, 155-163 (2009). 57. Kinney, J.M. Metabolic responses of the critically ill patient. Crit Care Clin 11, 569- 585 (1995). 58. Kominsky, D.J., Campbell, E.L. & Colgan, S.P. Metabolic shifts in immunity and inflammation. J Immunol 184, 4062-4068 (2010). 59. York, A.G., et al. Limiting Cholesterol Biosynthetic Flux Spontaneously Engages Type I IFN Signaling. Cell 163, 1716-1729 (2015). 60. Wu, D., et al. Type 1 Interferons Induce Changes in Core Metabolism that Are Critical for Immune Function. Immunity 44, 1325-1336 (2016). 61. Fritsch, S.D. & Weichhart, T. Effects of Interferons and Viruses on Metabolism. Front Immunol 7, 630 (2016). 62. Dubois, E.L. & Tuffanelli, D.L. Clinical Manifestations of Systemic Lupus Erythematosus. Computer Analysis of 520 Cases. JAMA 190, 104-111 (1964). 63. Jandacek, R.J., Heubi, J.E. & Tso, P. A novel, noninvasive method for the measurement of intestinal fat absorption. Gastroenterology 127, 139-144 (2004). 64. Gilham, D., et al. Carboxyl ester lipase deficiency exacerbates dietary lipid absorption abnormalities and resistance to diet-induced obesity in pancreatic triglyceride lipase knockout mice. The Journal of biological chemistry 282, 24642- 24649 (2007). 65. Palmer, C., Bik, E.M., DiGiulio, D.B., Relman, D.A. & Brown, P.O. Development of the human infant intestinal microbiota. PLoS biology 5, e177 (2007). 66. Costanzo-Garvey, D.L., et al. KSR2 is an essential regulator of AMP kinase, energy expenditure, and insulin sensitivity. Cell metabolism 10, 366-378 (2009). 67. Moreno-Fernandez, M.E., et al. Peroxisomal beta-oxidation regulates whole body metabolism, inflammatory vigor, and pathogenesis of nonalcoholic fatty liver disease. JCI Insight 3(2018). 68. Giles, D.A., et al. Thermoneutral housing exacerbates nonalcoholic fatty liver disease in mice and allows for sex-independent disease modeling. Nat Med 23, 829-838 (2017). 69. Harley, I.T., et al. IL-17 signaling accelerates the progression of nonalcoholic fatty liver disease in mice. Hepatology 59, 1830-1839 (2014). 70. Quarta, C., et al. Molecular Integration of Incretin and Glucocorticoid Action Reverses Immunometabolic Dysfunction and Obesity. Cell Metab 26, 620-632 e626 (2017). 71. Pfuhlmann, K., et al. Celastrol-Induced Weight Loss Is Driven by Hypophagia and Independent From UCP1. Diabetes 67, 2456-2465 (2018). 72. Kabra, D.G., et al. Hypothalamic leptin action is mediated by histone deacetylase 5. Nat Commun 7, 10782 (2016). 73. Pfluger, P.T., et al. Calcineurin Links Mitochondrial Elongation with Energy Metabolism. Cell Metab 22, 838-850 (2015). 74. Cappelletti, M., et al. Type I interferons regulate susceptibility to inflammation- induced preterm birth. JCI Insight 2, e91288 (2017).

137

75. Ogata, H., et al. The toll-like receptor protein RP105 regulates lipopolysaccharide signaling in B cells. J Exp Med 192, 23-29 (2000). 76. Fischer, K., et al. Alternatively activated macrophages do not synthesize catecholamines or contribute to adipose tissue adaptive thermogenesis. Nat Med 23, 623-630 (2017). 77. Sanchez-Gurmaches, J., et al. Brown Fat AKT2 Is a Cold-Induced Kinase that Stimulates ChREBP-Mediated De Novo Lipogenesis to Optimize Fuel Storage and Thermogenesis. Cell Metab 27, 195-209 e196 (2018). 78. Mukherjee, R., et al. Nicotinamide adenine dinucleotide phosphate (reduced) oxidase 2 modulates inflammatory vigor during nonalcoholic fatty liver disease progression in mice. Hepatol Commun 2, 546-560 (2018). 79. Bray, N.L., Pimentel, H., Melsted, P. & Pachter, L. Near-optimal probabilistic RNA- seq quantification. Nat Biotechnol 34, 525-527 (2016). 80. Chen, J., Bardes, E.E., Aronow, B.J. & Jegga, A.G. ToppGene Suite for gene list enrichment analysis and candidate gene prioritization. Nucleic Acids Res 37, W305-311 (2009). 81. O'Rourke, R.W., Gaston, G.D., Meyer, K.A., White, A.E. & Marks, D.L. Adipose tissue NK cells manifest an activated phenotype in human obesity. Metabolism 62, 1557-1561 (2013). 82. Tschop, M.H., et al. A guide to analysis of mouse energy metabolism. Nat Methods 9, 57-63 (2011).

138

ACKNOWLEDGEMENTS

This study was supported, in part, by American Diabetes Association (ADA) 1-18-IBS-

100 (to S.D.); CCHMC Pediatric Diabetes and Obesity Center (to S.D., M.H., and T.H.I.);

NIH R01DK099222 (to S.D.); NIH R01DK099222-02S1 (associated with S.D. and

M.E.M.F.); American Heart Association (AHA) 17POST33650045 (to M.E.M.F.); ADA 1-

19-PMF-019 (to M.E.M.F.); R01AI075159 (to C.L.K.); NIH T32AI118697 and

T32GM063483-14 (associated with C.C.C.); Albert J. Ryan Foundation Fellowship (to

I.T.W.H.); NIH HD07463 and GM063483 (associated with I.T.W.H.); and PHS Grant P30

DK078392 Pathology of the Digestive Disease Research Core Center at CCHMC

(associated with S.D.). Marie Skłodowska Curie training network “ChroMe” grant H2020-

MSCA-ITN-2015-675610 (to M.H.T. and P.T.P.); German Center for Diabetes Research

(DZD) (to M.H.T. and P.T.P.); Initiative and Networking Fund of the Helmholtz Association

(to M.H.T. and P.T.P.); Helmholtz-Israel-Cooperation in Personalized Medicine (to

P.T.P.); Helmholtz Initiative for Personalized Medicine (iMed) (to M.H.T.); and Helmholtz

Portfolio Program “Metabolic Dysfunction” (to M.H.T.). We thank S. Burden for providing

Mlc1f-cre mice. The Cd180tm1a(KOMP)Wtsi targeted ES cells used for this research project were generated by the trans-NIH Knock-Out Mouse Project (KOMP) and obtained from the KOMP Repository (www.komp.org). We thank Daniel Giles, Matthew Lawson, Jarren

Oates, Jessica Doll, Pablo Alarcon and the CCHMC Pediatrics Diabetes and Obesity

Center team for technical assistance.

AUTHOR CONTRIBUTIONS

C.C.C., I.T.W.H., P.T.P, A.T., T.E.S., J.L.A., M.E.M.-F, M.S.M.A.D. participated in data generation. C.C.C., I.T.W.H., P.T.P, A.T., T.E.S., J.L.A., M.E.M.-F., M.S.M.A.D., M.J.F.,

139

L.M.F, J.S.-G., R.K., M.H., T.H.I., S.P.W., S.J.P., D.A.R., R.J.S., M.H.T., C.L.K., and S.D. participated in data analysis, interpretation, provided materials and technical support and participated in review of the manuscript. C.C.C., I.T.W.H., C.L.K., and S.D. participated in the conception and design of the study and wrote the manuscript.

COMPETING FINANCIAL INTERESTS

S.D., C.L.K and J.L.A hold patents on BAFF and APRIL. M.H.T. is a scientific advisor to

Novo Nordisk and ERX.

140

FIGURE LEGENDS

Figure 1. BAFF axis regulates obesity development. (a) Weight gain of WT or RP105-

/- mice fed a high-fat diet (HFD) for 24 weeks. (b-d) WT, RP105-/- or RP105-/-/BAFF-/- mice were fed a HFD for 16 weeks. (b) Weight gain. (c) Fasting glucose. (d) Glucose tolerance test (GTT). (e-f) WT, RP105-/-, µMT or BAFF-Tg mice fed a HFD for 20 weeks. (e)

Systemic BAFF concentration. (f) Mean weight gain. (g-j) WT or BAFF-Tg mice fed a

HFD for 20 weeks. (g) Fasting glucose at 20 weeks. (h) GTT at 12 weeks. (i) Liver triglycerides. (j) Systemic alanine transaminase (ALT) at 20 weeks. (a) Representative of

6 independent experiments, n = 4-8/condition. (b-d) Representative of 3 independent experiments, n = 6/condition. (e-f) Representative of 3 independent experiments, n = 6-

8/condition. (g-j) Representative of 3 independent experiments, n = 6/condition. (a-b, d, h) Area under the curve. (c, e-f) One-way ANOVA. (g, i-j) Student’s t-test. *p < 0.05, **p

< 0.01, ***p < 0.001, ****p < 0.0001.

Figure 2. BAFF modifies white adipose lipid handling. (a) White adipocytes stimulated in the presence or absence of recombinant BAFF (rBAFF; 500 ng/ml) and subjected to

RNA-seq analysis. Ontology pathways and heat map of selected associated genes. (b) eWAT mRNA expression of indicated lipolysis genes in WT, RP105-/- or BAFF-Tg mice after 20 weeks on a HFD. (c-d) White adipocytes treated with saline (NS) or rBAFF (500 ng/ml) for 24 hours. (c) Free glycerol. (d) mRNA expression of indicated lipolysis genes.

(e) CD-fed WT mice were treated i.p. with rBAFF (2 µg/mouse) every other day for one week. eWAT mRNA expression of indicated lipolysis genes. (f) Free glycerol in supernatants of WT, BAFF-R-/-, TACI-/- or BCMA-/- adipocytes treated with saline (NS) or

141 rBAFF (500 ng/ml). (a) A single experiment, n = 2/condition. (b) Representative of 3 independent experiments, n = 6-8/condition. (c-d) Representative of 3 independent experiments, n = 3-4/condition. (e-f) Representative of 2 independent experiments, n =

3-4/condition. (g) Representative of 2 independent experiments, n = 3-4/condition. Data represents mean +/- SEM. (b-f) Student’s t-test. *p < 0.05, **p < 0.01, ***p < 0.001.

Figure 3. BAFF modifies brown adipose adaptive thermogenesis. (a-b) WT, RP105-

/- or BAFF-Tg mice were fed a HFD for 20 weeks. (a) BAT UCP-1 mRNA expression. (b)

BAT mitochondria oxygen consumption. (c-d) Brown adipocytes were treated with saline

(NS) or rBAFF (500 ng/ml) for 6 hours. (c) UCP-1 mRNA expression. (d) Oxygen consumption rate (OCR). (e) Oxygen consumption of WT mice treated with saline (NS) or rBAFF (2 µg/mouse) for 24 hours prior to norepinephrine (NE; 1mg/kg) challenge. (f- g) Obese WT mice treated with rBAFF (2 µg/mouse) every other day for one week and monitored in TSE Phenomaster. (f) Monitoring of energy expenditure over 5 days. (g) Bar graph of combined energy expenditure. (a-b) Representative of n = 6-8/condition. (c-d)

Representative of 3 independent experiments, n = 3-4/condition. (e) Representative of 2 independent experiments, n = 3-4/condition. (f-g) A single experiment, n = 3-6/condition.

Data represents mean +/- SEM. (a-d) Student’s t-test. (e) Area under the curve. (f-g) analysis of covariance (ANCOVA) with body weight as covariate. *p < 0.05, ***p < 0.001.

Figure 4. APRIL modifies white adipose lipolysis and brown adipose adaptive thermogenesis. (a) Weight gain of WT or BAFF-/- mice fed a HFD for 20 weeks. (b)

Systemic APRIL concentration in WT or BAFF-/- mice. (c-d) White adipocytes stimulated

142 in the presence or absence of rBAFF (500 ng/ml), rAPRIL (500 ng/ml), or saline and subjected to RNA-seq analysis. (c) Venn diagram analysis of genes modified by BAFF and APRIL. (d) Ontology pathways and heat map of associated overlapped genes. (e) mRNA expression of indicated lipolysis genes in white adipocytes treated with saline (NS) or rAPRIL (500 ng/ml) for 24 hours. (f) Lean WT mice treated with rAPRIL (2 µg/mouse) every other day for two weeks. eWAT mRNA expression of indicated lipolysis genes. (g)

Free glycerol in supernatants of WT, BAFF-R-/-, TACI-/- or BCMA-/- adipocytes treated with saline (NS) or rAPRIL (500 ng/ml) for 24 hours. (h-i) WT brown adipocytes treated with saline (NS) or rAPRIL (500 ng/ml) for 6 hours. (h) UCP-1 mRNA expression. (i) Oxygen consumption rate. (a-b) Representative of 3 independent experiments, n = 5-6/condition.

(c-d) A single experiment, n = 2/condition. (e-f) Representative of 3 independent experiments, n = 3/condition. (g) Representative of 2 independent experiments, n = 3-

5/condition. (h) Representative of 3 independent experiments, n = 3/condition. (h) A single experiment, n = 3-5/condition. Data represents mean +/- SEM. (a-b, e-i) Student’s t-test. *p < 0.05.

Figure 5. APRIL-/- mice are protected from obesity development. (a-h) WT and

APRIL-/- mice were fed a HFD for 20 weeks. (a) Systemic BAFF concentration. (b) Weight gain. (c) Food Intake (d) Energy expenditure at 16 weeks of HFD. (e) Fasting glucose at

20 weeks of HFD. (f) GTT at 14 weeks of HFD. (g) Fasting insulin at 20 weeks of HFD.

(h) Systemic ALT at 20 weeks of HFD. (i) Weight gain of WT, BAFF-/-, APRIL-/- or BAFF-

/-/APRIL-/- mice fed a HFD for 20 weeks. (a-h) Representative of 3 independent experiments, n = 6-7/condition. (i) A single experiment, n = 4-6/condition. Data represents

143 mean +/- SEM. (a, c-e, g-h) Student’s t-test. (b, f, i) Area under the curve. *p < 0.05, **p

< 0.01, ****p < 0.0001.

Figure 6. Effect of BAFF and APRIL axis is conserved in human adipocytes. (a-b)

Human primary adipocytes isolated from persons undergoing bariatric surgery.

Adipocytes were treated with Saline (NS), rBAFF (500 ng/ml), or rAPRIL (500 ng/ml) for

24 hours as indicated. (a) Supernatant free glycerol. (b) LIPE mRNA expression normalized to NS. (c) PNPLA2 mRNA expression normalized to NS. (d) BAFF and (e)

APRIL systemic concentration in lean or obese persons. (f) BAFF and (g) APRIL systemic concentration correlation with BMI. (h) BAFF and (i) APRIL systemic concentration correlation with change in BMI 1-year after bariatric surgery. (a-c) Representative persons, n = 10-12/condition. (d-g) Data combined, n = 4 lean and n = 30 obese. (h-i)

Data combined, n = 22 persons undergoing bariatric surgery. Data represents mean +/-

SEM. (a-e) Student’s t-test. (f-i) Linear regression. *p < 0.05, ****p < 0.0001.

Figure 7. BAFF and APRIL axis is a regulator of adipose tissue/adipocyte homeostasis and body weight gain. A proposed model of the impact of the

BAFF/APRIL axis on white and brown AT/adipocyte lipid handling. BAFF/APRIL regulation of AT/adipocytes function is associated with protection from DIO in mice. The effect of this axis positively correlates with greater loss of body weight in bariatric patients.

144

145

146

147

148

149

150

151

MATERIALS AND METHODS

Mouse Obesogenic Diet Model

All experiments utilized male mice on a C57BL/6 background (Jackson). WT, RP105-/-,

RP105flox, Vav1cre, CD19cre, Nestincre, MLCcre, Villincre, RP105-/-/BAFF-/-, μMT, BAFF-Tg,

APRIL-/-, BAFF-/-/APRIL-/- mice were bred at Cincinnati Children’s Hospital Medical Center

(CCHMC) in a specific pathogen-free (spf) facility maintained at 22˚C, with free access to autoclaved chow diet food (CD; LAB Diet #5010; calories provided by carbohydrates

[58%], fat [13%] and protein [29%]) and water. Mice, 6-8 weeks of age, were fed either an irradiated high-fat diet (HFD; Research Diets #D12492i; 60% of calories from fat) or a

CD and food was replaced weekly. For energy absorption studies, fecal pellets were collected at 15 weeks and analyzed by bomb calorimetry. For lipid absorption studies, at

24 weeks, mice were switched to Behenate mouse diet and fecal pellets were collected at day 3 and day 4 for fat absorption analysis63,64. Phylogenetic analysis of the fecal microbiota was performed essentially as previously described via 16S rRNA pyrosequencing of fecal pellets collected after 8 weeks of HFD feeding65. Body composition and in vivo metabolic phenotyping were performed as previously described66-

69. Prior to glucose metabolism testing, insulin tolerance testing, or terminal harvest mice were fasted overnight as previously described66-69. Briefly, fasted mice were subjected to glucose (mice injected 100 µL of a 10% dextrose solution per gram of body weight) or insulin (mice injected with 100 µL of a 0.15 units/mL solution of insulin [Novolin] per gram of body) tolerance was determined. Energy expenditure was quantified by Phenomaster

(TSE systems)67,70-73. Systemic BAFF (R&D systems) and APRIL (MyBioSource) levels were quantified as per manufacturer instructions. All care was provided in accordance

152 with the Guide for the Care and Use of Laboratory Animals. All studies were approved by the CCHMC IACUC.

Reciprocal Bone Marrow Transfer

Bone marrow from WT or RP105-/- mice were transferred into RP105-/- recipient mice as previously described67,74. Briefly, femur-derived bone marrow cells (5 × 106) were transferred into whole-body irradiated RP105-/- recipient mice. Flow cytometric analysis of peripheral blood chimerism was performed 10 weeks after bone marrow reconstitution prior to mice being placed on a HFD or Chow diet.

Generation of RP105Flox/Flox line.

An ES Cell clone (D11) containing the conditionally targeted allele, Cd180tm1a(KOMP)Wtsi, of

Cd180, the gene encoding RP105, was obtained from the knockout mouse project

(KOMP). This clone passed quality control and was selected for blastocyst injection. ES cells were grown on neomycin resistant mouse embryonic fibroblast feeders (a gift of M.

Flick) and were then injected into C57BL/6–Albino blastocysts (B6(Cg)-Tyrc-2J/J stock

#00058 from Jackson Labs http://jaxmice.jax.org/strain/000058.html) and implanted into pseudopregnant C57BL/6 mothers by the CCHMC transgenic mouse core. These embryo transfers yielded numerous chimeric males of high chimerism, several of which were capable of transmitting the targeted allele in the germline as confirmed by PCR.

The germline-targeted progeny of these chimeras were subsequently bred to

FLPe-Deleter mice1 (B6.Cg-Tg(ACTFLPe)9205Dym/J stock #005703 from Jackson Labs http://jaxmice.jax.org/strain/005703.html ) to excise the FRT-flanked knockout-first β-

153 galactosidase reporter/selection cassette2. Excision of the cassette was confirmed by

PCR for both the neomycin selection cassette and with primers (IH152 and IH154) flanking the FRT cassette. This PCR product was sequenced to confirm the presence of both the FRT and loxP sites. For colony maintenance, the presence of the Exon 3 distal loxP site was confirmed by PCR using primers complementary to the genomic DNA flanking the loxP site (IH123 and IH124). This product was also sequenced to confirm the presence of the loxP site. The Rp105Flox/WT mice thus generated were then bred to

C57BL/6NJ mice. Progeny lacking the FlpE-deleter transgene (confirmed via PCR) were selected to establish an Rp105Flox/Flox breeding colony after breeding to homozygosity for both the Floxed allele and a wild-type Nnt allele3. Genotyping of RP105Flox/Flox mice was carried out with the following primer pairs:

IH123 Forward: CCATCTGAGAAAGAAGAGCATTTACC,

IH124 Reverse: TGAGCATTAGATTTTGCTGGGAC;

IH152 Forward: GCATTTCCCCATCTATCATCTGAC,

IH154 Reverse: TATTGCTAACATCGTCCGCCTAC.

Exon 3 containing the majority of the coding sequence of RP105 (Cd180), was identified as a likely to result in knockout by a prediction algorithm used by KOMP and was successfully targeted in the generation of the germline RP105 knockouts75.

Mouse Primary Adipocytes

Isolation and digestion (1mg/ml Collagenase Type IV, Dispase 2, CaCl2) of inguinal white

AT was as previously described76. Preadipocytes within stromal vascular fraction (SVF) were cultured until confluence and differentiated as previously described76. Briefly,

154 initiation media (Growth media [DMEM:F12, FBS, Pen/Strep], Rosiglitazone,

Dexamethasone, 3-Isobutyl-1-methylxanthine, insulin) was utilized for 2 days, afterwards cells were changed to continuation media (Growth media, Rosiglitazone, Insulin) for 2 days and followed by differentiation media (Growth media, Insulin) for an additional 2 days. Differentiated white adipocytes were utilized for downstream processes.

Brown preadipocytes were isolated and differentiated as previously described77.

Briefly, brown preadipocytes were isolated from P1 B6/J neonates, maintained in high- glucose DMEM in incubators at 37˚C and 5% CO2 and immortalized with pBabe-SV40

Large T. The gender of neonates was not determined. Brown preadipocytes were subjected to differentiation media (high-glucose DMEM including 10% FBS, 1% antibiotics, 20nM insulin and 1nM T3) brown preadipocyte. After 4 days, cells were switched to induction media (high-glucose DMEM including 10% FBS, 1% antibiotics, insulin 20nM, 1nM T3, 0.125mM indomethacin, 2 µg/mL dexamethasone and 0.5mM 3- isobutyl-1-methylxanthine (IBMX) for 2 days. Following this, differentiation media was utilized until day 12.

Adipocyte free glycerol quantification

Murine or Human primary adipocytes were cultured in the presence or absence of BAFF

(500ng/ml) or APRIL (500ng/ml) for 24 hours. Free glycerol was quantified by colorimetric assay (Sigma Aldrich) as per manufacturer instructions.

qRT-PCR

155

Adipocytes were homogenized in TRIzol (Invitrogen) followed by RNA extraction, reverse transcription to cDNA (Verso cDNA Synthesis Kit; Thermo Scientific) and qPCR analysis

(Light Cycler 480 II; Roche) – according to manufacturer’s instruction as previously described67-69,74,78.

The following primers pairs were used for mouse studies: Bactin For

GGCCCAGAGCAAGAGAGGTA Rev GGTTGGCCTTAGGTTTCAGG – LIPE For

TCTCGTTGCGTTTGTAGTGC Rev ACGCTACACAAAGGCTGCTT – PNPLA2 For

GTTGAAGGAGGGATGCAGAG Rev GCCACTCACATCTACGGAGC – UCP1 For

TCAGCTGTTCAAAGCACACA Rev GTACCAAGCTGTGCGATGTC – F4/80 For

CTTTGGCTATGGGCTTCCAGTC Rev GCAAGGAGGACAGAGTTTATCGTG – CD68

For CTTCCCACAGGCAGCACAG Rev AATGATGAGAGGCAGCAAGAGG – TNF For

CCAGACCCTCACACTCAGATCA Rev CACTTGGTGGTTTGCTACGAC – IL1b For

GGTCAAAGGTTTGGAAGCAG Rev TGTGAAATGCCACCTTTTGA – IFNg For

TGGCTGTTTCTGGCTGTTACTG Rev ACGCTTATGTTGTTGCTGATGG – CCL2 For

TGTCTGGACCCATTCCTTCTTG Rev AGATGCAGTTAACGCCCCAC – CCL3 For

ACCATGACACTCTGCAACCAAG Rev TTGGAGTCAGCGCAGATCTG – IL6 For

TGGTACTCCAGAAGACCAGAGG Rev AACGATGATGCACTTGCAGA – CXCL1 For

ACCCAAACCGAAGTCATAGC Rev TCTCCGTTACTTGGGGACAC. mRNA expression of each gene was compared to Actb (beta-actin, mouse) expression.

RNA sequencing and gene expression quantification

Gene expression of primary white adipocytes was determined as previously described79 by running 50 base pair single-end reads (~30 million reads per sample). Following the

156 removal of barcodes and primers, raw reads were aligned to the mm10 genome using kallisto80, which quantifies transcript abundances of high-throughput sequencing reads.

Kallisto pseudoaligns reads to a reference, by identifying transcripts that are compatible with each raw read, with annotations provided by UCSC. The psuedoalignment generates accurate quantification, with transcripts per million (TPM) as the output. All further processing and analyses were performed in GeneSpring 14.9 GX. Each transcript was log2-normalized and baselined to the median across all samples. Reasonably expressed transcripts (raw TPM >3 in 100% of samples in at least one condition) were included for differential analysis. Differential expression was determined through unpaired t-tests with an FDR-corrected p-value cutoff of 0.05 and a fold change requirement of >2. For pathway analysis, the database at toppgene.cchmc.org was employed, which amasses ontological data from over 30 individual repositories. RNA-sequencing raw data can be accessed at GSE131298.

Cellular Bioenergetics Quantification

Primary brown adipocytes (1x104/well) or BAT-derived mitochondria were plated in a polyethylenimine pre-coated XF24 Cell culture microplate as previously described67.

Briefly, a XF Analyzer (Seahorse Bioscience) was used to measure bioenergetics. Briefly, a XF24 extracellular flux assay cartridge (Seahorse Bioscience) was hydrated overnight at 37˚C according to manufacturer’s instruction. MAS-1 buffer supplemented with sucrose

(70mM), D-mannitol (200mM), KH2PO4 (5mM), Mg2Cl (5mM), HEPES (2mM), EDTA

(1mM) and 0.2% fatty free acid BSA (pH 7.4) was incubated at 37˚C in a non-CO2 incubator for 1 hour. Pyruvate/malate (30mM), GDP (10mM), FCCP (40µM), antimycin

157

(40µM) were sequentially injected and cellular oxygen consumption rate (OCR) was quantified.

Human Subjects

Lean or bariatric surgery participants were recruited from the Cincinnati Children’s

Hospital Medical Center (CCHMC) Pediatric Diabetes and Obesity Center. Exclusion criteria included alcohol abuse, viral and autoimmune hepatitis, immunosuppressive or steroid use. Clinical phenotypes at time of surgery and one-year post-surgery is provided in Supplemental Table 1. Recruitment and study protocols were approved by the institutional review board at CCHMC.

Human Primary Adipocytes

Isolation and digestion of omental white adipose tissue (40mg/ml Type II Collagenase) collected at the time of surgery was as previously described protocols81. Filtration of digested tissue was subsequently centrifuged at 250g and subjected to Ack lysis buffer to isolate the SVF. SVF was cultured in expansion media (DMEM/F:12, 15% FBS, 1%

Pen-strep) until confluence and subjected to human adipocyte differentiation media

(DMEM/F:12, 1% Pen-strep, 2mM glutamine, 15mM HEPES, 10mg/ml transferrin, 33µM biotin, 0.5µM insulin, 17µM pantothenate, 0.1µM dexamethasone, 2nM T3, 500µM IBMX,

1µM ciglitazone) for 14-16d, followed by 7-10d in human adipocyte maintenance media

(DMEM/F:12, 1% Pen-strep, 2mM glutamine, 15mM HEPES, 10mg/ml transferrin, 33µM biotin, 0.5µM insulin). Differentiated adipocytes were utilized for downstream processes.

158

Human Systemic Cytokine Quantification

Human BAFF or APRIL plasma concentrations from lean or persons with severe obesity were determined by ELISA using MilliplexTM Multiplex kits (MilliporeSigma) according to manufacturer’s protocol. Briefly, 25µL of plasma, plated in duplicate on a 96 well black plate, was incubated with 25µL of antibody coated beads. Plates were subsequently washed and 25µL of secondary antibody was incubated, then by 25µL of streptavidin-

RPE. 150µL of sheath fluid was added to plates that were washed and then read using luminex technology on a Milliplex Analyzer (milliporeSigma). Data analysis performed by the Cincinnati Children’s Medical Center Research Flow Cytometry Core.

Statistical analysis

Statistical tests were utilized for all data sets with similar variance. Choice of test was dependent on number of groups and whether normal distribution exists. For all normally distributed data Student’s t-test was used for 2 groups whereas one-way ANOVA was utilized for 3 or more groups with Tukey’s post hoc test to determine differences between groups. Collective indirect calorimetry data was analyzed using analysis of covariance (ANCOVA) with body weight as covariates.82 All data presented as means

+/- SEM. P values < 0.05 were considered significant. Analysis was performed via

GraphPad Prism Software’s. Determined sample sizes were based on preliminary data with respect to obesity modeling including weight gain, immune cell infiltration, severity of obesity-associated sequelae, and interrogation of myeloid cell inflammatory vigor. No animals were excluded from the analyses and none of the studies were blinded.

159

160

161

162

163

164

165

166

167

168

169

SUPPLEMENTAL FIGURE LEGENDS

Supplemental Figure 1. Intrinsic lack of RP105 expression does not explain effects on protection from obesity pathogenesis. RP105-/- or WT mice were fed a CD or HFD for 24 weeks. (a) Nutrient and (b) fat absorption at 16 weeks on HFD. (c) Body weight after 4 doses of leptin or vehicle control administration at 6 weeks on HFD. (d) Phylum composition of intestinal microbiota in WT and RP105-/- placed on a CD or HFD. (a-d)

Representative of 3 independent experiments, n = 6-8/condition. Data represents mean

+/- SEM. (a-c) Student’s t-test.

Supplemental Figure 2. Non-hematopoietic RP105 expression is the locus of protective effects from obesity development. (a) Scheme of reciprocal bone marrow transfer. (b) Weight gain of RP105-/- and WT-/- recipients fed a HFD for 20 weeks. (c)

Scheme of RP105 flox mouse design. (d-g) RP105 conditional knockout mice fed a HFD for 12 weeks. (d) Deletion in total immune cells (Vav1cre). (e) Deletion in B cells (CD19cre).

(f) Deletion in skeletal muscle (MLCcre). (g) Deletion in central and peripheral nervous system (Nestincre). (a-b) Representative of 2 independent experiments, n = 6/condition.

(d-g) Representative of 2 independent experiments, n = 4-8/condition. Data represents mean +/- SEM. (b, d-g) Student’s t-test.

Supplemental Figure 3. Deletion of BAFF in RP105-/- mice reverts fat pad distribution. RP105-/-/BAFF-/- and WT mice were fed a HFD for 16 weeks and WAT tissue distribution was analyzed. Representative of 3 independent experiments, n = 4-

6/condition. Data represents mean +/- SEM. Student’s t-test. *p < 0.05, ** p < 0.01.

170

Supplemental Figure 4. μMT mice are protected from obesity-associated glucose dysmetabolism. WT and μMT mice were fed a HFD for 20 weeks. (a) Fasting glucose at 20 weeks on HFD. (b) GTT at 12 weeks on HFD. Representative of 4 independent experiments, n = 4-6/condition. Data represents mean +/- SEM. (a) Student’s t-test. (b)

Area under the curve. *p < 0.05, **p < 0.01, ***p < 0.001.

Supplemental Figure 5. BAFF’s effects on lipolysis and UCP-1 induction are specific to white and brown adipocytes respectively. White and brown adipocytes were treated with saline (NS) or rBAFF (500 ng/ml) for 24 hours and 6 hours respectively.

(a) Free glycerol. (b) UCP-1 mRNA expression. Representative of 2 independent experiments, n = 3-4/condition. Data represents mean +/- SEM. (a-b) Student’s t-test. *p

< 0.05.

Supplemental Fig. 6. Exogenous rBAFF administration augments energy expenditure. Lean CD-fed WT mice treated with rBAFF (2 µg/mouse) every other day for one week and monitored in TSE Phenomaster. (a) Energy expenditure prior to treatment. (b) Energy expenditure in light cycle. (c) Energy expenditure in dark cycle. (d)

RER. (e) Body Weight one-week post-injections. Representative of 2 independent experiments, n = 4-6/condition. Data represents mean +/- SEM. (a-c) analysis of covariance (ANCOVA) with body weight as covariate. (d-e) Student’s t-test. Student’s t- test. *p < 0.05.

171

Supplemental Figure 7. BAFF-/- mice exhibit increased body weight. (a-c) WT or

BAFF-/- mice fed a HFD for 20 weeks. (a) Body weight. (b) Fasting glucose at 13 weeks on HFD. (c) Systemic ALT at 20 weeks of HFD. Representative of 3 independent experiments, n = 5-6/condition. Data represents mean +/- SEM. Student’s t-test. *p <

0.05.

Supplemental Fig. 8. Exogenous rAPRIL administration in lean mice augments energy expenditure. (a-e) Lean CD-fed WT mice were injected with rAPRIL (2

µg/mouse) every other day for one week and monitored in TSE Phenomaster. (a) Energy expenditure prior to treatment. (b) Energy expenditure in light and dark cycles. (c) RER.

(d) Body Weight one-week post-injections. (e) Obese HFD-fed (20 weeks) WT mice were injected with rAPRIL (2 µg/mouse) every other day for one week and monitored in TSE

Phenomaster. Energy expenditure in light and dark cycles. (a-d) Representative of 2 independent experiments, n = 4-6/condition. (e) A single experiment, n = 4-6/condition.

Data represents mean +/- SEM. (a-b, e) Analysis of covariance (ANCOVA) with body weight as covariate. (c-d) Student’s t-test.

Supplemental Fig. 9. APRIL does not induce inflammatory programming in adipocytes. (a) White adipocytes stimulated in the presence or absence of rAPRIL (500 ng/ml) for 24 hours and subjected to RNA-seq analysis. (a) APRIL-specific ontology pathways and heat maps of associated genes. (b-e) mRNA expression of indicated cytokines and chemokines. (a) A single experiment, n = 2/condition. (b-e) Representative

172 of 3 independent experiments, n = 3/condition. Data represents mean +/- SEM. (b-e)

Student’s t-test. *p < 0.05.

Supplemental Table 1. Characteristics of the cohort undergoing bariatric surgery.

173

Chapter 5. Conclusions, Discussion and Future Directions

Adipocytes were long perceived to be simple and inert cells. This dissertation demonstrates a clear need to redefine our understanding of adipocyte effector capacity by highlighting a significant underappreciation for adipocytes’ role within the inflammatory milieu and consequent contribution to human health and disease. While therapeutic approaches targeting inflammation, including in obesity-associated metabolic disease1, have seen success, some approaches have not reached their predicted optimal therapeutic potential2,3. These therapeutic approaches were developed from evidence primarily stemming from observations of immune cell contributions to inflammation. Thus, these approaches may not account for the inflammatory contribution of the non- hematopoietic compartment, including adipocytes. Our work presented in this dissertation suggests that adipocyte-inflammation axis is a highly capable player in the maintenance of health or as potential drivers of disease. A greater understanding of the interplay between inflammation and adipocyte biology may yield novel insights into more selective and potentially attractive therapies. Data through this dissertation provide insights into the dichotomy of both detrimental and beneficial aspects of the adipocyte-inflammatory axis including the type I IFN/IFNAR and BAFF/APRIL axes. Hence, careful attention will need to be given in future approaches tilting this delicate balance. Collectively, this dissertation has begun to shed direct light on the mechanisms regulating the interplay between adipocytes and inflammation. In this chapter key findings and suggestions for future and deeper definition of mechanisms underlying inflammatory regulation of adipocyte biology will be discussed.

174

TYPE I IFN/IFNAR CONTROL OF ADIPOCYTE INFLAMMATORY VIGOR

A. Type I IFN/IFNAR axis mechanisms controlling adipocyte inflammatory potential

Our data demonstrate that the type I IFN/IFNAR axis uncovers an “immune-like” gene signature in adipocytes, akin to type I IFN primed macrophages (Chapter 3). Adipocyte

IFNAR sensing further unleashes a glycolysis-associated inflammatory vigor in mice and humans, demonstrating a functional convergence to macrophages. Notably, inhibition of glycolysis via the glucose analog 2-deoxyglucose (2-dg), reverses type I IFN driven inflammatory vigor in adipocytes. However, the underpinning mechanisms regulating this adipocyte-intrinsic inflammation remain undefined. Our data indicates that type I IFNs augment adipocyte glycolysis and increase expression of both hypoxia-inducible factor 1- a (HIF-1a) and pyruvate kinase (PKM2), central determinants of aerobic glycolysis4,5.

HIF-1a induces PKM2, in turn PKM2 can also activate HIF-1a creating a positive feedback loop5,6. Hence, a salient question invoked by our data is whether type I IFN licensing of glycolysis-associated inflammation in adipocytes is driven by key mediators of the flux through glycolysis (e.g., HIF-1a, PKM2). As type I IFN regulation of glucose metabolism is a key to its antiviral activity7, if HIF-1a and/or PKM2 mediates type I IFN- driven adipocyte-intrinsic inflammatory contribution to obesity pathogenesis or to other type I IFN-driven diseases (e.g., SLE, Infection [Listeria monocytogenes]) is similarly unknown.

Future Direction

To begin probing these questions, we should initially define the necessity of HIF-1a or

Pkm2 in adipocyte-intrinsic inflammatory vigor by leveraging adipocyte-specific HIF-1a

(AdipoqcreHIF1afl/fl) deficient or Pkm2 (AdipoqcrePkm2fl/fl) deficient mice. We would

175 anticipate that adipocytes isolated from HIF-1a or Pkm2-deficient mice would exhibit blunted IFNb augmentation of ECAR and OCR (quantified using Seahorse technology), suggesting HIF-1a and Pkm2 are key regulators of IFNb-mediated aerobic glycolysis in adipocytes. In addition, HIF-1a or Pkm2-deficient adipocytes would exhibit reduced

IFNb+LPS-driven inflammatory vigor (e.g., IL-6, TNF). We would also expect that: (a) quantitative PCR (qPCR) of type I IFN axis signature genes, including Oas1a and Isg15, would be damped in HIF-1a or Pkm2-deficient adipocytes; (b) western blotting (WB) for key IFNAR signaling mediators (e.g., Jak1, STAT1, pSTAT1, IRF9) would be decreased.

In addition, to examine the relation between HIF-1a and Pkm2 in type I IFN/IFNAR-driven effects in adipocytes, we would also determine in the presence or absence of IFNb: (a)

HIF-1a expression (qPCR and WB) in Pkm2-deficient adipocytes and (b) Pkm2 expression (qPCR and WB) in HIF-1a- deficient adipocytes. As HIF-1a and PKM2 partake in a positive feedback loop, we would predict that lack of HIF-1a or PKM2 would dampen expression of the reciprocal mediator. We should also pursue the functional contribution of HIF-1a or Pkm2-driven adipocyte-specific inflammatory vigor to disease including employing HFD feeding (obesity model) or an infection model (e.g., Listeria, influenza A) in AdipoqcreHIF1afl/fl or AdipoqcrePkm2fl/fl mice. Collectively, formal definition of the role of

HIF-1a and Pkm2 in uncovering adipocyte inflammatory vigor would strengthen our insights into relationship between adipocyte metabolism, inflammatory vigor and contribution to pathology.

Metabolism, transcriptome, and epigenome form a complex interplay to modulate inflammatory responses.8-11 Our findings demonstrate that the type I IFN/IFNAR axis drives a convergence in adipocyte transcriptome and epigenome towards macrophages 176

(Chapter 3). Notably, type I IFNs modify macrophage epigenome to uncover their inflammatory vigor.12 As type I IFNs enhance adipocyte aerobic glycolysis, HIF-1a, and

PKM2 expression, whether HIF-1a or PKM2 are critical mediators of type I IFN-driven transcriptome and epigenome programming is unknown. To answer this series of questions, IFNb primed or saline treated adipocytes from Adipoq-Cre-/HIF1afl/fl, Adipoq-

Cre-/PKM2fl/fl (controls) or Adipoq-Cre+/HIF1afl/fl, Adipoq-Cre+/PKM2fl/fl mice would be submitted for RNA-seq (differential gene expression, transcriptional networks) and ATAC- seq (local chromatin accessibility). Initial analyses would focus on the expression of key type I IFN signature and inflammatory genes (Ifb1, Irf9, Oas1a, Isg15, Il6, Tnf, Hif1a,

Pkm2). These studies would uncover additional novel mechanistic targets in HIF-1a and

Pkm2 regulation of IFNb-driven adipocyte inflammatory vigor. The potential reciprocal relation between HIF1a and PKM2 in IFNb primed adipocytes would also be revealed. If

HIF1a and PKM2 is found to be a critical regulator of type I IFN/IFNAR driven effects in mouse adipocytes, use of pharmacological of HIF-1a (Echinomycin; R&D systems)13 or

PKM2 modulators (inhibitor III [inhibitor; Calbiochem]14, DASA-10 [activator]15) in human adipocytes could be employed. CRISPR/Cas9 has been used successfully in human primary cells16-18 and could be used to delete HIF-1a or PKM2 in human adipocytes. In sum, these studies would provide additional mechanistic insights into the relationship between type I IFN-driven core metabolism, transcriptome and epigenome to modify inflammatory vigor in adipocytes.

B. Inflammatory modulation of adipocyte core metabolism

177

Adipocytes are highly metabolic cells. We determined that the type I IFN/IFNAR axis amplifies aerobic glycolysis in adipocytes (Chapter 3). Inflammation modifies immune cell metabolomics and metabolomic profiles are suggested to be potentially valuable predictive and preventive markers of disease19-23. Profound alterations to the metabolome may be linked with development of obesity-associated sequelae24. The causative agents triggering the transition from a metabolically healthy to a metabolically challenged state in persons with severe obesity remains poorly defined. The type I IFN signature is higher in obese Met-C people (Chapter 3). Whether the type I IFN/IFNAR modifies the metabolomic landscape of adipocytes is unknown. These findings could be potentially exploited and used as predictive and preventative markers to identify persons with obesity at risk for development of metabolic derangements.

Future Direction

To shed light on type I IFN/IFNAR alteration of adipocyte metabolome, WT and IFNAR-/- adipocytes, stimulated with saline, LPS, IFNb, IFNb+LPS, would be analyzed by liquid chromatography mass spectroscopy (LCMS) to identify the alterations to metabolic substrates. Significantly altered intermediates can be tested for functional relevance in type I IFN/IFNAR augmentation of adipocyte inflammatory vigor. For instance, if glutamate was identified to be significantly increased by type I IFN activation, cultures using glutamate free media can be used in IFNb/LPS studies. Pharmacological modulators of the enzymes that convert these metabolites into intermediates used for metabolic processes (e.g., Glycolysis, TCA cycle) could be employed to further test the necessity of identified metabolites in adipocyte type I IFN/IFNAR axis effects.

178

The type I IFN signature is increased in persons with obesity and metabolic derangements (Chapter 3). Given this, it would be highly pertinent to define the systemic metabolomic profile of these cohorts and to correlate identified metabolites with the type

I IFN signature. Further, adipocytes from persons with obesity (metabolically challenged and health) would be treated with saline, LPS, IFNb, IFNb+LPS and analyzed by LCMS.

Findings here could provide potential novel clinical markers of the transition between metabolically healthy and challenged states in persons with severe obesity.

C. Identifying the Relevant Source(s) of Type I IFN in Obesity

Obesity augments the type I IFN signature across various organs, including AT, liver, and spleen (Chapter 3). Adipocytes produce IFNb in response to a TLR challenge and obesity amplifies adipocyte IFNb expression. In addition to adipocytes, various immune cells that infiltrate AT (e.g., pDCs, macrophages) are well-established producers of type I IFN25.

Hence, several pertinent questions arise including: (a) what is the relevant source of type

I IFN within the AT in obesity?; and (b) are type I IFNs derived from a local, within the AT, and/or systemic source in obesity? Preliminary evidence suggests that systemic type I

IFNs can be sensed by AT. Specifically, exogenous recombinant IFNb, Lymphocytic choriomeningitis virus (LCMV; inducer of type I IFN), or a gut-tropic Clostridium difficile a b c 2.5 Ctrl 40 LCMV 100 C.diff challenges are sufficient to induce rIFNβ 2.0 30 **** 80 **** 1.5 60 WAT IFNb expression (Fig. 1). 20 1.0 40 Figure 1. Exogenous IFNb or pathogen 0.5 10 20 challenge is sufficient to induce WAT IFNb Arbitrary Units (AU) 0.0 0 0 mRNA expression. WT mice were challenged with IFNB1 IFNB1 IFNB1 (a) exogenous IFNb (104U/mouse), (b) Lymphocytic choriomenigitis virus (LCMV; 5d post-infection), or (c) Clostiridum difficile (C. diff; 3d post-infection) and mRNA expression of IFNB1 was quantified in WAT; normalized to respective Ctrl. Representative 2 independent experiments, n = 3-4/condition. Mean +/- SEM. ****P < 0.0001.

179

Future Direction

To define the primary contributors, and temporal boundaries, of IFNb in the context of obesity, IFNb-YFP reporter mice (IFNmob) would be placed on a CD or HFD. Kinetic analysis (every 4 weeks) of IFNb producing cell type(s) in PBMC, spleen, WAT

(epididymal, inguinal and perirenal) and liver tissues from lean and obese IFNbmob mice would be examined via: (a) histological quantification of IFNb-YFP expression by fluorescent microscopy26; (b) FACS of IFNb-YFP in immune cells including neutrophils, macrophages, dendritic cells; (c) qPCR analysis of type I IFN axis (e.g., IFNb, IFNAR1,

IRF9, Oas1a, ISG15) expression in adipocytes, hepatocyte and immune cells. As IFNb conditional knockout mice do exist, as an initial approach we would begin to identify the compartment (hematopoietic or non-hematopoietic) contributing to type I IFN production in obesity, we would perform reciprocal bone marrow transfers, complemented with clodronate liposomes treatment to deplete radioresistant resident macrophages, between

WT (CD45.1) and IFNb-/- (CD45.2) mice. Mice would be subsequently challenged with

HFD or CD and kinetically analyzed (every 2 weeks) as above. Although definition of the source of IFNa would be also pertinent, mice consists of 13 different IFNa subtypes and thus formal definition of contributors to IFNa would be challenging.

Our findings suggest that type I IFN effects are conserved in humans including: (a) the type I IFN signature is increased in Met-C obese persons and (b) adipocytes from

Met-C persons exhibit increased responsiveness to type I IFN-driven inflammatory vigor.

To define the source(s) of type I IFN in AT of obese patients, AT biopsies from Met-H and

Met-C bariatric patients would be subjected to an immunofluorescence approach, staining using Abs for IFNβ, IFNAR, Adipoq, CD45, CD11b, CD11c and F4/80. This would identify

180 if IFNβ is produced by adipocytes and/or immune cells (e.g., macrophages) and where

IFNAR expression is within the AT milieu. In turn, this information would be correlated with clinical parameters to determine if differences exist between Met-H obese and Met-

C obese persons. These insights would yield an improved understanding of the type I

IFN/IFNAR axis in persons with severe obesity, with particular emphasis on the transition between Met-H and Met-C obese states.

D. Temporality of type I IFN axis in adipocytes

Activation of TLRs is a key inducer of type I IFN production in both adipocytes (Chapter

3) and immune cells27. Type I IFNs partake in autocrine and paracrine loops to drive amplification of type I IFN production in immune cells.27,28 Temporal type I IFN production

Oligo FCCP propagates effects on functional capacity of 300 NS IFNβ immune cells, including alterations to core IFNβ + Eto 200 metabolism29. For instance, production of

phasic type I IFNs in pDCs is associated with 100 29 OCR (pmol/min) increasing fatty acid oxidation . However, the

0 0 10 20 30 40 50 60 70 temporal nature of type I IFN in adipocytes is Time (min) unexplored. Our findings begin to hint at the Figure 2. Etomoxir modifies maximal but not basal respiration in adipocytes. Adipocytes were treated in the presence or absence of IFNb (250U) or Etomoxir possibility of temporal effects of the type I IFN (250µM) as indicated. OCR measured by Seahorse analysis. Representative of 2 independent axis on adipocyte function. Specifically, while experiments, n = 6/condition. inhibition of mitochondrial β-oxidation, via Etomoxir, does not impact IFNβ-driven basal respiration in adipocytes, closer examination revealed that Etomoxir significantly dampened IFNβ-mediated, FCCP-driven maximal respiration (Fig. 2). Although Etomoxir

181 does not alter IFNβ-driven adipocyte inflammatory vigor (Chapter 3), further investigation to define if mitochondrial β-oxidation is a critical mediator of additional effector function(s) in adipocytes, including type I IFN production, is warranted. Critical knowledge of potential optimal timing of therapeutic interventions targeting adipocyte type I IFN release could be directly gained from these studies.

Future Direction

As a reductionist approach to define the autocrine/paracrine amplification of type I IFN production in adipocytes, WT and IFNAR-/- adipocytes (DCs in parallel as a control) would be stimulated IFNβ. IFNβ levels, measured via ELISA, would be performed 0hr, 4hr, 8hr,

12hr, 24hr post-stimulation. Given the “immune-like” similarities between adipocytes and myeloid cells, we would anticipate type I IFN amplification in WT, but not IFNAR-/-, adipocytes. Further, IL-6 and TNF can be measured to delineate whether potentiation of type I IFN production modulates type I IFN-driven adipocyte inflammatory vigor. To determine the positive feedback of type I IFN production on adipocyte bioenergetics, WT or IFNAR-/- adipocytes would be treated with IFNβ and subjected to Seahorse analysis at various time points (e.g., 0hr, 4hr, 8hr, 12hr, 24hr) post-stimulation to analyze OCR and

ECAR. Additionally, in vivo studies via kinetic measurements of systemic IFNβ levels and quantification of mRNA expression of IFNB1 in mature adipocytes after exogenous IFNβ administration would be performed. Of note, investigations of HFD-fed IFNβ-YFP reporter mice (discussed above) could also provide clues of whether obesity development drives phasic IFNβ production. If type I IFN effects on adipocytes are temporal, studies examining the impact of type I IFNs on adipocyte metabolome (discussed above) would be expanded to encompass multiple time points.

182

Suppressor of cytokine signaling (SOCS), including SOCS1 and SOCS3, are well- established counter-regulatory measures to persistent amplification of type I IFN production. Kinetic type I IFN induction likewise drives SOCS expression in immune cells.30,31 Despite canonical perception of SOCS as counter-regulatory measures to excessive inflammation, activation of SOCS impacts obesity-associated metabolic sequelae in both beneficial and detrimental aspects. Obesity augments AT SOCS1 and

SOCS3 expression32-34. Some reports demonstrate that SOCS1 and 3 in myeloid cells dampens their proinflammatory cytokine (e.g., IL-6, TNF) production and protects mice from HFD-driven insulin resistance35. In contrast, SOCS1 and 3 have implicated roles in the direct perturbation of insulin receptor signaling33 and glucose uptake, including in

AT32,36. Existing studies however have not examined the impact of SOCS in type I IFN- driven adipocyte responses. This exploration would be highly pertinent given that adipocytes are also insulin responsive and have the capacity to behave in an “immune- like” manner. Notably, administration of type I IFN in humans increases glucose dysmetabolism37. Hence, these studies would provide critical insights into the canonical counter-regulators of type I IFN/IFNAR axis on adipocyte metabolic function and may provide support of additional, tractable targets for obesity-associated sequelae.

To elucidate the role of SOCS1 and SOCS3 in type I IFN-driven adipocyte-intrinsic inflammation and insulin sensitivity, we would initially determine if type I IFN sensing in adipocytes augments SOCS1 and SOCS3 expression, via qPCR and WB. Given the similarities in type I IFN responsiveness between adipocytes and myeloid cells, we would likewise expect significant induction of both SOCS1 and SOCS3 in adipocytes. Next, adipocytes from AdipoqcreSOCS1fl/fl and AdipoqcreSOCS3fl/fl mice (SOCS1fl and SOCS3fl

183 available from Jackson) would be stimulated in the presence or absence of IFNb and/or

LPS. We anticipate that lack of SOCS1 or SOCS3 would remove the inhibitory inflammatory mechanisms and yield even greater adipocyte inflammatory vigor. In addition, we would examine insulin receptor signaling in IFNb treated WT, SOCS1-/- and

SOCS3-/- adipocytes through WB of the insulin receptor substrate 1 (IRS1) serine phosphorylation (inhibitory) and tyrosine phosphorylation (activation) subunits. We would also subject adipocyte-specific SOCS1- or SOCS3-deficient mice to daily rIFNb treatment and fasting glucose measurements. One week after daily rIFNb administration, we would perform an ITT to determine insulin sensitivity. Adipocytes would be isolated from mice determined to have insulin resistance and examined for inflammatory capacity and IRS1 subunit phosphorylation. Combined, these studies would determine the impact of SOCS1 and SOCS3 on type I IFN-driven adipocyte inflammation and insulin sensitivity.

E. Type I IFN/IFNAR axis control of adipokines

This dissertation highlights inflammatory mechanisms of controlling adipocyte functions

(e.g., inflammatory vigor, lipid handling). Whether inflammation more broadly impacts other aspects of adipocyte cellular function should be further explored. Adipocytes are well-established endocrine cells, capable of hormone production (e.g., leptin, adiponectin) that impact total body homeostasis38,39. Notably, AT is the primary producer of leptin38. Leptin resistance is a hallmark of obesity pathogenesis40. Therapeutic approaches targeting the leptin system in the context of obesity however have been less fruitful41. An intricate interplay exists between leptin and inflammatory cascades including bidirectional modulation between leptin and inflammatory mediators (e.g., TNF, IL-6)42,43.

184

Leptin, like IL-6, activates the STAT3 pathway44. Type I IFN/IFNAR axis augments IL-6 production and STAT3 pathway in adipocytes (Chapter 3). Our RNA-seq analysis of IFNb treated adipocytes indicates an induction of the leptin system and signaling by leptin

pathways (Fig. 3a). Congruently,

obese IFNAR-/- mice have

significantly decreased circulating

leptin levels (Fig. 3b). Non-

hematopoietic IFNAR expression

Figure 3. IFNb upregulates the leptin-associated pathways in contributes to obesity-associated adipocytes. (a) Adipocytes treated with saline or IFNb (250U) subjected to RNA-seq analyses. Ontology of upregulated pathways by IFNb in adipocytes and their associated genes. (b) Systemic leptin metabolic derangements (Chapter concentration, % change compared to WT. (a) A single experiment, n = 2/condition. (b) Representative of 3 independent experiments, n = 4-6/condition. Data represents Mean+/-SE. **p<0.01. 3). Hence, it is reasonable to posit that the type I IFN/IFNAR axis may similarly uncover the leptin system in adipocytes and further connect the type I IFN/IFNAR axis in adipocytes to global metabolic health.

Future Directions

To test the hypothesis that the type I IFN/IFNAR axis modifies the leptin production by adipocytes, we would stimulate WT and IFNAR-/- adipocytes with IFNb and/or LPS and measure leptin levels in the culture supernatants by ELISA. We anticipate that IFNb would be sufficient to increase leptin production while the combination of IFNb+LPS may further exacerbate leptin levels in WT, but not IFNAR-/-, adipocytes. WT and IFNAR-/- mice would also be treated with exogenous rIFNb and measurements encompassing systemic leptin levels and WAT leptin expression (qPCR and WB), would be performed. As leptin can induce production of inflammatory mediators42,43, we would also determine if leptin is sufficient and/or necessary to induce type I IFN production in adipocytes by treating WT

185 or LepSTAT3-/- (leptin-specific STAT3 signaling deficient) adipocytes with rLeptin and quantification of IFNb levels in the culture supernatants. Since obesity augments leptin production, if leptin does induce adipocyte IFNb levels this could suggest a potential mechanism by which obesity promotes IFNb levels and how leptin triggers IFNb-driven metabolic dysfunction during leptin sensitive stages of obesity development.

Our RNAseq data indicate that IFNb modulates leptin signaling in adipocytes and that obese IFNAR-/- exhibit decreased systemic leptin levels (Fig. 3). Notably, there remains a lack in clarity of the involvement of immune mechanisms in regulation of leptin sensitivity. As inflammation can also play beneficial roles in adipocytes (Chapter 4), whether type I IFN/IFNAR axis may hold some positive aspects to adipocyte function is an interesting possibility. To define whether the type I IFN/IFNAR axis modifies adipocyte leptin sensitivity, initial purview of STAT3 phosphorylation (Tyr705; subunit for leptin signaling) by WB analysis in IFNb and/or LPS treated adipocytes would be performed.

Further, we would administer rIFNb in WT or IFNAR-/- mice followed by a leptin challenge.

Responsiveness to leptin would be measured via body weight loss (intact leptin sensitivity drives weight loss). Combined, these approaches could unlock potentially novel avenues of an immune-mediated regulation of the leptin system.

F. Role of gut-adipocyte axis in tuning adipocyte inflammatory capacity.

AT/adipocytes are appreciated effectors of global metabolic health and that disruption to the delicate energy balance drives obesity. Obesity, in turn, shares a reciprocal relationship with the gut microbiome. Obesity alters the composition of gut microbiome45 and gut microbiota ecology is a central determinant of obesity pathogenesis46. The gut

186

microbiome likewise calibrates immune responses47.

Logically, studies have emerged highlighting a gut-

adipose tissue axis. However, the evidence of its impact

on adipose tissue function remains controversial.

Pathways activated by the gut microbiota, including

trimethylamine N-oxide (TMAO)48 are linked with

progression of obesity and dampening effects on WAT

beiging. In contrast, antibiotic depletion of gut microbiome

or utilization of germ-free mice impairs BAT and beige

thermogenesis49. The mechanisms of how the gut

communicates with AT to modify function remains Figure 4. Depletion of gut microbiome alters adipocyte inflammatory vigor. Adipocytes were elusive. Metabolites produced by gut microbes, including derived from mice treated saline (Ctrl), gram- and/or gram+ antibiotic cocktail butyrate49, have been proposed contributors to for 9 weeks. Adipocytes were treated in the presence or absence of LPS and supernatant TNF concentration was BAT/beige thermogenesis. However, whether the gut quantified. A single experiment, n = 1- 3/condition. microbiome impacts adipocyte inflammatory capacity is undefined and represents an intriguing possibility. Our preliminary data suggests that depletion of the microbiome with a gram-/gram+ cocktail alters adipocyte inflammatory vigor (Fig. 4). Notably, our primary adipocytes are derived from preadipocytes within the adipose tissue and are devoid of SVF. Maximal differentiation into adipocytes takes 12d.

This suggests that the gut microbiome may alter the epigenome of pre-adipocyte precursors that modify adipocyte inflammatory vigor. ATAC-seq of pre-adipocyte stem cells and differentiated adipocytes isolated from mice depleted of gut microbiota would provide key clues to this possibility.

187

BAFF/APRIL AXIS CONTROL OF ADIPOCYTE LIPID HANDLING

A. Additional BAFF/APRIL effects on lipid handling circuit.

Observations in this dissertation highlight a novel, NS rBAFF 2.5 2.0 rAPRIL potentially beneficial, means of immune-mediated 2.0 1.5 1.5 regulation of adipocyte lipid handling (Chapter 4). 1.0 1.0 0.5 0.5 Despite, BAFF/APRIL-driven augmentation of

Relative Expression 0.0 0.0 CD36 LPL lipolysis, we did not observe the metabolic

Figure 5. BAFF and APRIL induce brown dysfunction typically associated with increased adipocyte lipid uptake regulators. Brown adipocytes were treated with saline (NS), rBAFF (500ng/ml), or rAPRIL (500ng/ml). circulating lipids (e.g., glucose dysmetabolism) in all Indicated mRNA expression of lipid uptake regulators relative to NS. Representative of 2 independent experiments, n = 3/condition. of the strains of mice examined with increased circulating BAFF (i.e., RP105-/-, µMT, BAFF-Tg, and APRIL-/-). This suggests that evolutionary mechanisms may be in place to prevent accumulation of excess circulating lipids, mechanisms that may involve the BAFF/APRIL axis. Multiple mechanisms exist to handle circulating lipids including liver uptake (e.g., glycerol)50 and BAT (e.g., triglycerides)51. BAT uptake of circulating lipids can activate their thermogenic programming52. Our data suggests that BAFF and APRIL can induce critical mediators of brown adipocyte lipid uptake (e.g., CD36, Lipoprotein lipase [LPL]; Fig. 5). Of note,

BAFF/APRIL exerts effects on BAT/brown adipocyte thermogenesis (Chapter 4).

In addition to BAT lipid uptake, another explanation for the lack BAFF/APRIL- associated lipid-driven metabolic dysfunction is its potential impact on WAT beiging.

Beiging is associated with increased thermogenic activity (UCP-1 dependent53 and independent mechanisms54) and beneficial effects on global metabolic health55. Notably, white adipocytes utilized throughout this dissertation are derived from the inguinal WAT,

188 the fat pad most prominently associated with beiging. While, BAFF/APRIL did not induce

UCP-1 induction in our white adipocytes, BAFF, but not APRIL, did upregulate the sarcoplasmic/endoplasmic reticulum calcium ATPase (SERCA2b; regulator of UCP-1 independent thermogenesis; Fig. 6). Combined these data implore several salient questions including: (i) Does the BAFF/APRIL axis facilitate enhanced BAT lipid uptake? and (ii) Can the BAFF/APRIL axis promote WAT beiging and thermogenesis in a UCP-1 independent manner? Thus, in depth definition of the control of the BAFF/APRIL axis on the lipid handling circuit would yield additional targetable therapeutic avenues.

Figure 6. BAFF upregulates SERCA2b in adipocytes. SERCA2b mRNA expression in white adipocytes were treated in the presence or absence of rBAFF (500ng/ml) or rAPRIL (500ng/ml). Future Direction

To investigate BAFF/APRIL effects on BAT lipid uptake, assessment

of BAT CD36 and LPL mRNA and protein expression should be

performed in: (a) mice with increased circulating BAFF levels and (b) mice receiving exogenous rBAFF or rAPRIL. Lipid uptake studies would be performed wherein the aforementioned mice would receive radiolabeled lipids51,56 followed by subsequent monitoring of the presence of radiolabeled lipids within the BAT. As BAT lipid uptake activates their thermogenic programming, it is plausible that uptake of these exogenous lipids could be associated with BAT activation of UCP-1 and thermogenesis.

To better dissect the interlink of these two effects in the context of the BAFF/APRIL axis, it would be feasible to test this hypothesis by employing adipocyte-specific deletion

(Adipoqcre) of CD36 (CD36flox/flox; Jackson) and LPL (LPLflox/flox; Jackson). Here, we would anticipate that BAFF/APRIL administration to mice lacking BAT lipid uptake capacity would decrease UCP-1 activation and thermogenic activity.

189

Since our data suggests that BAFF/APRIL can upregulate the mRNA expression of SERCA2b but not UCP-1 in white adipocytes, it invokes the possibility of UCP-1 independent adipocyte beiging. Hence, a better definition of the extent to which the

BAFF/APRIL axis drives WAT beiging is required. Measurement of beige AT/adipocyte signature (e.g., PRDM16, CD137, TBX1)57 in the presence of BAFF or APRIL would be performed. It would be pertinent to determine if exogenous BAFF or APRIL treatment enhances the thermogenic capacity of the inguinal WAT by recording tissue temperature through the use of TSE technology.54,58 These approaches would provide a strong indication of BAFF and/or APRIL’s beiging potential. In parallel, adipocyte-specific

Serca2b-deficient mice51 could be utilized in these experiments to interrogate the potential interplay between the BAFF/APRIL axis and Serca2b to divulge a potential mechanism.

B. Mechanisms of BAFF/APRIL modulation of adipocyte lipid handling

This dissertation has unveiled a novel, potentially beneficial, role for the BAFF/APRIL axis in adipose tissue/adipocyte lipid handling. The collective impact of the BAFF/APRIL axis to modify obesity development (e.g., weight gain) underscores a potential clinical approach for weight management. Notably, BAFF and APRIL have broad effects including on adipocytes (Chapter 4), T cells59, and B cells60. Thus, therapeutic approaches exploiting the BAFF/APRIL system are likely to require specificity to prevent undesired off-target effects. Dissection of mechanisms that could determine BAFF/APRIL axis specificity includes: (i) definition of the critical receptor(s); and (ii) elucidation of underpinning molecular mechanisms on target tissues/cells.

Future Direction

190 a b The BAFF/APRIL axis is a highly *** ** 100 103 *** complex network which includes 3 80 * 102 60 receptors (BAFF-R/TACI/BCMA) 40 101

20 BAFF (ng/ml) 0 0 and 2 molecules (BAFF/APRIL). Mean weight gain (%) 10 -/- -/- -/- -/- -/- -/- WT WT TACI BCMA TACIBCMA BAFF-R BAFF-R This complexity likely stems from an c d e

600 evolutionary advantage to facilitate WT BAFF-R-/- TACI-/- BCMA-/- **** 250 500 400 200 400 300 specificity in sensing of 150 300 200 100 ALT (IU/L)

Glucose (mg/dl) *** 50 200 100 BAFF/APRIL in specific tissues. For

0 100 0 Fasting Glucose (mg/dl) 0 30 60 90 -/- -/- WT WT Time (min) instance, while each individual TACI-/- TACI BCMA-/- BAFF-R-/- BAFF-R BCMA -/-

Figure 7. WT, BAFF-R-/-, TACI-/- or BCMA-/- were placed on a HFD receptor is implicated to play for 24 weeks. (a) Mean weight gain. (b) Systemic BAFF levels. (c) Fasting glucose. (d) Glucose tolerance test). (e) Systemic ALT levels. Representative of 3 independent experiments, n = 4- different roles in B cell development 6/condition. Data represents Mean +/- SEM. **p < 0.01. ***p < 0.001. ****p < 0.0001. (e.g., BAFF-R necessary for B cell maturation, BCMA necessary for plasma cell survival), our data indicate that likely all 3 receptors are engaged on white adipocytes to activate lipolysis (Chapter 4). Consistent with these observations, deletion of a single receptor is insufficient to reverse resistance to HFD-driven weight gain (Fig. 7a). These single receptor knockouts exhibit enhanced systemic BAFF levels, likely as a compensatory effect (Fig. 7b). However, while lack of

TACI or BCMA protects mice from HFD-induced glucose dysmetabolism and hepatocellular damage, lack of BAFF-R does not drive similar effects (Fig. 7c-e).

Collectively this would suggest that better definition of the critical receptor(s) that drive adipocyte lipid handling would yield understandings of potential that could be harnessed in adipocyte-specific clinical approaches.

191

Lack of BAFF-R, TACI, or BCMA conditional knockout mice has hampered targeted dissection of the role of these receptors in the BAFF/APRIL axis. Generation of such mice would be highly valuable to the field. As an initial approach, mice with deletion of multiple receptors (e.g., BAFF-R-/-/TACI-/-, TACI-/-/BCMA-/-, BAFF-R-/-/TACI-/-/BCMA-/-) would be employed to elucidate their impact on obesity development through HFD feeding. All three receptors lie on different chromosomes, thus technical difficulties with the generation of mice consisting of multiple receptor deletion would not be anticipated.

Exogenous administration of BAFF or APRIL in these receptor deficient mice would expand on their role on WAT lipolysis and BAT thermogenesis. CRISPR/Cas9 technology, successfully employed in human primary cells16-18,61, can be utilized to generate deletion of these receptors in human adipocytes to ascertain if their effects are conserved. qPCR measurement of BAFF-R, TACI, and BCMA expression on human adipocytes would be performed in parallel. This expression data can be correlated to clinical characteristics of persons with severe obesity, with particular focus on body weight.

Beyond careful probing of the critical receptor(s) mediating BAFF/APRIL effects on adipocytes and obesity, examination of the underpinning molecular mechanisms may provide more clues to specificity of the BAFF/APRIL axis. Interferon regulatory factor 4

(IRF4), a transcriptional regulator of effector immune functions uniquely expressed in immune cells and adipocytes62,63, is a critical mediator of adipocyte lipid handling.

Targeted deletion of IRF4 in AT regulated DIO through alteration of lipid handling, including WAT lipolysis and BAT adaptive thermogenesis. However, how IRF4 is activated in adipocytes remains undefined. Our data suggests that WAT and BAT from

192

BAFF-Tg mice exhibit augmented mRNA expression of IRF4 (Fig. 8). BAFF was similarly sufficient to induce IRF4 expression in white adipocytes (Fig. 8). Given strong parallels in our observations of BAFF/APRIL axis and findings of IRF4 in adipocytes, we a b c WAT BAT Adipocytes hypothesize that IRF4 governs

WT WT BAFF/APRIL axis effects on adipocyte BAFF-Tg BAFF-R-/- 2.0 40 *** 10 * *** 1.5 30 8 lipid handling. Notably, we would employ 6 1.0 20 4 0.5 10 2 mice with adipocyte-specific IRF4 Relative Expression 0.0 Relative Expression 0 0 NS BAFF IRF4 IRF4 IRF4 (Relative Expression) deletion (AdipoqcreIRF4flox/flox) to Figure 8. BAFF axis increases IRF4 expression. (a-b) WT or BAFF-Tg mice fed a CD. (a) WAT or (b) BAT IRF4 interrogate this possible causal expression. (c) IRF4 expression of WT or BAFF-R-/- white adipocytes treated in the presence or absence of BAFF. (a-b) cre flox/flox Representative of 3 independent experiments, n = 4- relationship. Adipoq IRF4 mice 6/condition. (c) Representative of 3 independent experiments, n = 3/condition. Data represents Mean +/- SE. *p < 0.05. would be subjected to BAFF and/or

APRIL treatment to determine IRF4’s role in BAFF/APRIL driven energy expenditure. We would also transplant BAT from WT mice to AdipoqcreIRF4flox/flox mice followed by

BAFF/APRIL treatment. BAFF/APRIL administration would be expected to raise the temperature of WT transplanted BAT. Mature white adipocytes from AdipoqcreIRF4flox/flox would also be isolated and stimulated with BAFF or APRIL to define if IRF4 governs

BAFF/APRIL driven lipolysis. Further, as our data indicates that lack of BAFF-R abrogates

BAFF-driven induction of IRF4 (Fig. 7), interrogation of the critical receptor(s) driving this potential BAFF/APRIL interlink with IRF4 would similarly be of significant interest.

Complementing these findings, global unbiased approaches via RNA-seq and ChIP-seq would be performed to elucidate the molecular circuits in this setting--something that would yield additional novel mechanistic targets altering adipocyte function in this setting.

193

C. BAFF/APRIL axis impact on adipose tissue/adipocyte lipidomics

Lipids come in many different configurations (e.g., short-chain fatty acids [SCFA], long- chain fatty acids [LCFA], very long-chain fatty acids [VLCFA], saturated/unsaturated fatty acids). Clear links between lipidomic landscape, inflammation and disease pathogenesis exist including obesity64, NAFLD65, infection66, autoimmunity67, pregnancy complications68 and cancer69. Although controversial, saturated fatty acids (SFA) are believe to be pro-inflammatory and capable of directly activating TLRs on immune cells or non-hematopoietic cells (e.g. adipocytes)70. Myeloid cells, including macrophages, express multiple, functional, lipid uptake receptors including CD36, Lox1 and ABCA1.71

Macrophage activation of fatty acid synthase is proposed to skew them towards a proinflammatory M1 phenotype, while induction of fatty acid oxidation skews macrophages towards an M2 state71. Notably, the BAFF/APRIL axis has the capacity to amplify adipocyte lipolysis and release of lipids (Chapter 4). However, how BAFF/APRIL axis shapes the adipocyte lipidome is unknown. If BAFF/APRIL control of the adipocyte lipidome potentially fuels beneficial or detrimental effects in disease is an exciting possibility to provide additional novel predictive, preventative, and therapeutic targets.

Future direction

To begin to define how the BAFF/APRIL axis modifies the adipocyte lipidome, LCMS, a comprehensive and specific method of analysis, should be performed on the lipid fraction.

This global view would allow for a better definition of BAFF/APRIL-driven shifts in balance between lipids. To more carefully delineate the relevance of identified shifts in lipid homeostasis, mouse models with genetic inhibition of those lipid biosynthesis pathways would be employed (e.g., SREBP Cleavage-Activating Protein [SCAP]-/-). We would

194 predict that alteration to lipid biosynthesis pathways would alter BAFF/APRIL-driven effects on lipolysis and adipocyte inflammation. Exogenous administration of BAFF and/or

APRIL, in WT or BAFF-R-/-, TACI-/-, or BCMA-/- mice, would likewise yield useful observations of the relevance of lipid homeostasis to BAT-mediated energy expenditure.

195

TYPE I IFN/IFNAR AND BAFF-APRIL BALANCE IN ADIPOCYTE INFLAMMATION

The stories presented in this dissertation underscore two inflammatory pathways that can spur adipocytes to augment their inflammatory vigor (Type I IFN/IFNAR axis) or induce adipocyte lipid handling machinery (BAFF/APRIL axis). However, a balance between the type I IFN/IFNAR and BAFF/APRIL axes likely exists in parallel to facilitate adipocyte homeostasis. Notably, the type I IFN and BAFF/APRIL axes are interconnected. Type I

IFNs can induce the production of BAFF or APRIL in various hematopoietic72,73 and non- hematopoietic cells74,75. Functionally, lack of BAFF abrogates type I IFN driven murine models of SLE-like disease76 and blockade of type I IFN dampens BAFF expression77.

How the type I IFN/IFNAR and BAFF/APRIL axes are interlinked in adipocytes is unexplored, but fully warrants investigation. For instance, we postulate that activation of both pathways in adipocytes could be beneficial in the context of infections to accelerate clearance of a pathogen via: (a) augmentation of adipocyte inflammation; (b) release of lipids as fuel for immune cells; and (c) activation of adaptive thermogenesis to increase core body temperature. Type I IFNs alters lipid biosynthesis to activate innate immune responses against viral infections66. If these type I IFN-driven effects are mediated via

BAFF/APRIL is unknown. In contrast, it is also conceivable that type I IFN and

BAFF/APRIL axes activation in adipocytes could contribute to the development and/or exacerbation of autoimmune diseases (e.g., Lupus, Sjögren's Syndrome).

Future Directions

We would leverage mice that modulate either type I IFN/IFNAR (IFNAR-/-, IFNARfl/fl) or

BAFF/APRIL (BAFF-/-, APRIL-/-, BAFF-R-/-, TACI-/-, BCMA-/-) axes to more carefully dissect apart the connection between these two pathways in adipocytes. A preliminary

196 experiment to determine whether these pathways are potentially related in adipocytes would be to measure type I IFN levels in BAFF or APRIL stimulated adipocytes and conversely to examine BAFF or APRIL levels in type I IFN treated adipocytes.

Additionally, employment of BAFF or APRIL treatment in adipocyte-specific IFNAR deleted mice in models of autoimmunity (e.g., Lupus-like models [NZB/W, Pristane]) or infection (e.g., Influenza A, Liseteria monocytogenes) would provide clues to the relevance of the intersection between these pathways in adipocytes.

197

SUMMARY: OVERARCHING POTENTIAL EXPLOITATION OF ADIPOCYTE-

INFLAMMATION AXIS IN HUMAN HEALTH AND DISEASE

Data presented throughout this dissertation underscore a vast underappreciation for the contribution of adipocytes as capable players within the inflammatory milieu. While this dissertation has naturally veered towards the impact of adipocyte inflammation obesity development and pathogenesis, our findings are suggestive of the wide potential impact of adipocyte-inflammation axis in maintenance of human health and disease (Fig. 9).

Adipocytes draw several parallels to immune cell counterparts in their inflammatory function. However, findings in this dissertation merely scratch the surface of our understanding of this adipocyte-inflammatory axis. As immune responses and inflammatory pathways are well characterized in immune cells, an exciting opportunity presents itself to leverage this understanding to maximize knowledge of adipocyte inflammatory capacity. Unification of our novel findings in adipocytes with this enormous existing literature could yield a plethora of new, attractive, and adipocyte-targeted preventative and therapeutic approaches to augment health or stem disease.

198

Figure 9. Adipocyte-intrinsic inflammatory axis: An underappreciated contributor to human health and disease. In spite of a long-held reputation as simple energy storing cells, adipocytes (e.g., white and brown adipocytes) are highly complex biosynthetic factories now appreciated to have multipotent effects across total body homeostasis. Although physiologically important interactions exist between inflammation, immune responses, and adipocyte biology, mechanisms underpinning inflammatory-mediated regulation of adipocyte function are poorly understood. This dissertation has begun to shed direct light of two inflammatory pathways that impact adipocyte effector function: (a) uncovering dormant adipocyte inflammatory potential (Type I IFN/IFNAR Axis); and (b) modulation of adipocyte lipid handling

(BAFF/APRIL axis). Although this dissertation has highlighted the potential relevance of immune-mediated regulation of adipocyte function in the context of obesity, findings here may be broadly applicable across other type I IFN/IFNAR or BAFF/APRIL-driven diseases including cancer, neurological disease, autoimmunity, and infectious disease. As such, deeper investigation of adipocyte-intrinsic inflammatory axis spawns many exciting, bench to bedside opportunities. Congruently, more nuanced understanding of the balance between the type I IFN/IFNAR and BAFF/APRIL axes in adipocyte function may uncover more optimal bedside strategies and previously unappreciated clinical targets.

199

REFERENCES

1. Goldfine, A.B. & Shoelson, S.E. Therapeutic approaches targeting inflammation for diabetes and associated cardiovascular risk. J Clin Invest 127, 83-93 (2017). 2. Onuora, S. Obesity hampers effects of anti-TNF agents. Nat Rev Rheumatol 14, 320 (2018). 3. Gupta-Ganguli, M., Cox, K., Means, B., Gerling, I. & Solomon, S.S. Does therapy with anti-TNF-alpha improve glucose tolerance and control in patients with type 2 diabetes? Diabetes Care 34, e121 (2011). 4. Israelsen, W.J. & Vander Heiden, M.G. Pyruvate kinase: Function, regulation and role in cancer. Semin Cell Dev Biol 43, 43-51 (2015). 5. Cheng, S.C., et al. mTOR- and HIF-1alpha-mediated aerobic glycolysis as metabolic basis for trained immunity. Science 345, 1250684 (2014). 6. Luo, W., et al. Pyruvate kinase M2 is a PHD3-stimulated coactivator for hypoxia- inducible factor 1. Cell 145, 732-744 (2011). 7. Burke, J.D., Platanias, L.C. & Fish, E.N. Beta interferon regulation of glucose metabolism is PI3K/Akt dependent and important for antiviral activity against coxsackievirus B3. J Virol 88, 3485-3495 (2014). 8. Das, S.K. Integrating transcriptome and epigenome: putting together the pieces of the type 2 diabetes pathogenesis puzzle. Diabetes 63, 2901-2903 (2014). 9. Phan, A.T., Goldrath, A.W. & Glass, C.K. Metabolic and Epigenetic Coordination of T Cell and Macrophage Immunity. Immunity 46, 714-729 (2017). 10. Wong, C.C., Qian, Y. & Yu, J. Interplay between epigenetics and metabolism in oncogenesis: mechanisms and therapeutic approaches. Oncogene 36, 3359-3374 (2017). 11. Novakovic, B., et al. beta-Glucan Reverses the Epigenetic State of LPS-Induced Immunological Tolerance. Cell 167, 1354-1368 e1314 (2016). 12. Park, S.H., et al. Type I interferons and the cytokine TNF cooperatively reprogram the macrophage epigenome to promote inflammatory activation. Nat Immunol 18, 1104-1116 (2017). 13. Wang, Y., Liu, Y., Malek, S.N., Zheng, P. & Liu, Y. Targeting HIF1alpha eliminates cancer stem cells in hematological malignancies. Cell Stem Cell 8, 399-411 (2011). 14. Vander Heiden, M.G., et al. Identification of small molecule inhibitors of pyruvate kinase M2. Biochem Pharmacol 79, 1118-1124 (2010). 15. Anastasiou, D., et al. Pyruvate kinase M2 activators promote tetramer formation and suppress tumorigenesis. Nat Chem Biol 8, 839-847 (2012). 16. Chu, H.W., et al. CRISPR-Cas9-mediated gene knockout in primary human airway epithelial cells reveals a proinflammatory role for MUC18. Gene Ther 22, 822-829 (2015). 17. Schumann, K., et al. Generation of knock-in primary human T cells using Cas9 ribonucleoproteins. Proc Natl Acad Sci U S A 112, 10437-10442 (2015). 18. Liang, X., et al. Rapid and highly efficient mammalian cell engineering via Cas9 protein transfection. J Biotechnol 208, 44-53 (2015). 19. Beger, R.D. A review of applications of metabolomics in cancer. Metabolites 3, 552-574 (2013).

200

20. Jutley, G.S. & Young, S.P. Metabolomics to identify biomarkers and as a predictive tool in inflammatory diseases. Best Pract Res Clin Rheumatol 29, 770-782 (2015). 21. Santoru, M.L., et al. Cross sectional evaluation of the gut-microbiome metabolome axis in an Italian cohort of IBD patients. Sci Rep 7, 9523 (2017). 22. Shommu, N.S., et al. The Use of Metabolomics and Inflammatory Mediator Profiling Provides a Novel Approach to Identifying Pediatric Appendicitis in the Emergency Department. Sci Rep 8, 4083 (2018). 23. Young, S.P., et al. The impact of inflammation on metabolomic profiles in patients with arthritis. Arthritis Rheum 65, 2015-2023 (2013). 24. Cirulli, E.T., et al. Profound Perturbation of the Metabolome in Obesity Is Associated with Health Risk. Cell Metab 29, 488-500 e482 (2019). 25. Gonzalez-Navajas, J.M., Lee, J., David, M. & Raz, E. Immunomodulatory functions of type I interferons. Nat Rev Immunol 12, 125-135 (2012). 26. Bauer, J., et al. Cutting Edge: IFN-beta Expression in the Spleen Is Restricted to a Subpopulation of Plasmacytoid Dendritic Cells Exhibiting a Specific Immune Modulatory Transcriptome Signature. J Immunol 196, 4447-4451 (2016). 27. Trinchieri, G. Type I interferon: friend or foe? J Exp Med 207, 2053-2063 (2010). 28. Marie, I., Durbin, J.E. & Levy, D.E. Differential viral induction of distinct interferon- alpha genes by positive feedback through interferon regulatory factor-7. EMBO J 17, 6660-6669 (1998). 29. Wu, D., et al. Type 1 Interferons Induce Changes in Core Metabolism that Are Critical for Immune Function. Immunity 44, 1325-1336 (2016). 30. Piganis, R.A., et al. Suppressor of cytokine signaling (SOCS) 1 inhibits type I interferon (IFN) signaling via the interferon alpha receptor (IFNAR1)-associated tyrosine kinase Tyk2. J Biol Chem 286, 33811-33818 (2011). 31. Ivashkiv, L.B. & Donlin, L.T. Regulation of type I interferon responses. Nat Rev Immunol 14, 36-49 (2014). 32. Emanuelli, B., et al. SOCS-3 inhibits insulin signaling and is up-regulated in response to tumor necrosis factor-alpha in the adipose tissue of obese mice. J Biol Chem 276, 47944-47949 (2001). 33. Ueki, K., Kondo, T. & Kahn, C.R. Suppressor of cytokine signaling 1 (SOCS-1) and SOCS-3 cause insulin resistance through inhibition of tyrosine phosphorylation of insulin receptor substrate proteins by discrete mechanisms. Mol Cell Biol 24, 5434- 5446 (2004). 34. Ueki, K., Kondo, T., Tseng, Y.H. & Kahn, C.R. Central role of suppressors of cytokine signaling proteins in hepatic steatosis, insulin resistance, and the metabolic syndrome in the mouse. Proc Natl Acad Sci U S A 101, 10422-10427 (2004). 35. Sachithanandan, N., et al. Macrophage deletion of SOCS1 increases sensitivity to LPS and palmitic acid and results in systemic inflammation and hepatic insulin resistance. Diabetes 60, 2023-2031 (2011). 36. Shi, H., Tzameli, I., Bjorbaek, C. & Flier, J.S. Suppressor of cytokine signaling 3 is a physiological regulator of adipocyte insulin signaling. J Biol Chem 279, 34733- 34740 (2004). 37. Koivisto, V.A., Pelkonen, R. & Cantell, K. Effect of interferon on glucose tolerance and insulin sensitivity. Diabetes 38, 641-647 (1989).

201

38. Park, H.K. & Ahima, R.S. Physiology of leptin: energy homeostasis, neuroendocrine function and metabolism. Metabolism 64, 24-34 (2015). 39. Rosen, E.D. & Spiegelman, B.M. Adipocytes as regulators of energy balance and glucose homeostasis. Nature 444, 847-853 (2006). 40. Lee, J.H., Reed, D.R. & Price, R.A. Leptin resistance is associated with extreme obesity and aggregates in families. Int J Obes Relat Metab Disord 25, 1471-1473 (2001). 41. Santoro, A., Mattace Raso, G. & Meli, R. Drug targeting of leptin resistance. Life Sci 140, 64-74 (2015). 42. Mueller, W.M., et al. Evidence that glucose metabolism regulates leptin secretion from cultured rat adipocytes. Endocrinology 139, 551-558 (1998). 43. Shen, J., Sakaida, I., Uchida, K., Terai, S. & Okita, K. Leptin enhances TNF-alpha production via p38 and JNK MAPK in LPS-stimulated Kupffer cells. Life Sci 77, 1502-1515 (2005). 44. Buettner, C., et al. Critical role of STAT3 in leptin's metabolic actions. Cell Metab 4, 49-60 (2006). 45. Ley, R.E., et al. Obesity alters gut microbial ecology. Proc Natl Acad Sci U S A 102, 11070-11075 (2005). 46. Turnbaugh, P.J., et al. An obesity-associated gut microbiome with increased capacity for energy harvest. Nature 444, 1027-1031 (2006). 47. Round, J.L. & Mazmanian, S.K. The gut microbiota shapes intestinal immune responses during health and disease. Nat Rev Immunol 9, 313-323 (2009). 48. Schugar, R.C., et al. The TMAO-Producing Enzyme Flavin-Containing Monooxygenase 3 Regulates Obesity and the Beiging of White Adipose Tissue. Cell Rep 19, 2451-2461 (2017). 49. Li, B., et al. Microbiota Depletion Impairs Thermogenesis of Brown Adipose Tissue and Browning of White Adipose Tissue. Cell Rep 26, 2720-2737 e2725 (2019). 50. Landau, B.R., et al. Glycerol production and utilization in humans: sites and quantitation. Am J Physiol 271, E1110-1117 (1996). 51. Bartelt, A., et al. Brown adipose tissue activity controls triglyceride clearance. Nat Med 17, 200-205 (2011). 52. Fedorenko, A., Lishko, P.V. & Kirichok, Y. Mechanism of fatty-acid-dependent UCP1 uncoupling in brown fat mitochondria. Cell 151, 400-413 (2012). 53. Harms, M. & Seale, P. Brown and beige fat: development, function and therapeutic potential. Nat Med 19, 1252-1263 (2013). 54. Ikeda, K., et al. UCP1-independent signaling involving SERCA2b-mediated calcium cycling regulates beige fat thermogenesis and systemic glucose homeostasis. Nat Med 23, 1454-1465 (2017). 55. Bartelt, A. & Heeren, J. Adipose tissue browning and metabolic health. Nat Rev Endocrinol 10, 24-36 (2014). 56. Lynes, M.D., et al. Cold-Activated Lipid Dynamics in Adipose Tissue Highlights a Role for Cardiolipin in Thermogenic Metabolism. Cell Rep 24, 781-790 (2018). 57. Wu, J., et al. Beige adipocytes are a distinct type of thermogenic fat cell in mouse and human. Cell 150, 366-376 (2012). 58. Meyer, C.W., Ootsuka, Y. & Romanovsky, A.A. Body Temperature Measurements for Metabolic Phenotyping in Mice. Front Physiol 8, 520 (2017).

202

59. Mackay, F. & Leung, H. The role of the BAFF/APRIL system on T cell function. Semin Immunol 18, 284-289 (2006). 60. Mackay, F., Schneider, P., Rennert, P. & Browning, J. BAFF AND APRIL: a tutorial on B cell survival. Annu Rev Immunol 21, 231-264 (2003). 61. Xue, R., et al. Clonal analyses and gene profiling identify genetic biomarkers of the thermogenic potential of human brown and white preadipocytes. Nat Med 21, 760- 768 (2015). 62. Eguchi, J., et al. Transcriptional control of adipose lipid handling by IRF4. Cell Metab 13, 249-259 (2011). 63. Eguchi, J., et al. Interferon regulatory factors are transcriptional regulators of adipogenesis. Cell Metab 7, 86-94 (2008). 64. Hernandez-Carretero, A., et al. Obesity-induced changes in lipid mediators persist after weight loss. Int J Obes (Lond) 42, 728-736 (2018). 65. Moreno-Fernandez, M.E., et al. Peroxisomal beta-oxidation regulates whole body metabolism, inflammatory vigor, and pathogenesis of nonalcoholic fatty liver disease. JCI Insight 3(2018). 66. York, A.G., et al. Limiting Cholesterol Biosynthetic Flux Spontaneously Engages Type I IFN Signaling. Cell 163, 1716-1729 (2015). 67. Bernard, N.J. IL-27 links lipids to autoimmunity. Nat Rev Rheumatol 14, 385 (2018). 68. Smith, C.J., et al. Maternal dyslipidemia and risk for preterm birth. PLoS One 13, e0209579 (2018). 69. Long, J., et al. Lipid metabolism and carcinogenesis, cancer development. Am J Cancer Res 8, 778-791 (2018). 70. Glass, C.K. & Olefsky, J.M. Inflammation and lipid signaling in the etiology of insulin resistance. Cell Metab 15, 635-645 (2012). 71. Remmerie, A. & Scott, C.L. Macrophages and lipid metabolism. Cell Immunol 330, 27-42 (2018). 72. Panchanathan, R. & Choubey, D. Murine BAFF expression is up-regulated by estrogen and interferons: implications for sex bias in the development of autoimmunity. Mol Immunol 53, 15-23 (2013). 73. Banchereau, J. & Pascual, V. Type I interferon in systemic lupus erythematosus and other autoimmune diseases. Immunity 25, 383-392 (2006). 74. Lundell, A.C., et al. IFN type I and II induce BAFF secretion from human decidual stromal cells. Sci Rep 7, 39904 (2017). 75. Ittah, M., et al. Viruses induce high expression of BAFF by salivary gland epithelial cells through TLR- and type-I IFN-dependent and -independent pathways. Eur J Immunol 38, 1058-1064 (2008). 76. Jacob, N., et al. B Cell and BAFF dependence of IFN-alpha-exaggerated disease in systemic lupus erythematosus-prone NZM 2328 mice. J Immunol 186, 4984- 4993 (2011). 77. Sjostrand, M., et al. The Expression of BAFF Is Controlled by IRF Transcription Factors. J Immunol 196, 91-96 (2016).

203

APPENDIX

204

ARTICLES

Thermoneutral housing exacerbates nonalcoholic fatty liver disease in mice and allows for sex-independent disease modeling

Daniel A Giles1,2, Maria E Moreno-Fernandez1, Traci E Stankiewicz1, Simon Graspeuntner3, Monica Cappelletti1, David Wu4, Rajib Mukherjee1, Calvin C Chan1,2, Matthew J Lawson1, Jared Klarquist1,2, Annika Sünderhauf5, Samir Softic6, C Ronald Kahn6, Kerstin Stemmer7, Yoichiro Iwakura8, Bruce J Aronow9, Rebekah Karns10, Kris A Steinbrecher10, Christopher L Karp11, Rachel Sheridan12, Shiva K Shanmukhappa12, Damien Reynaud13, David B Haslam14, Christian Sina5, Jan Rupp3, Simon P Hogan4 & Senad Divanovic1

Nonalcoholic fatty liver disease (NAFLD), a common prelude to cirrhosis and hepatocellular carcinoma, is the most common chronic liver disease worldwide. Defining the molecular mechanisms underlying the pathogenesis of NAFLD has been hampered by a lack of animal models that closely recapitulate the severe end of the disease spectrum in humans, including bridging hepatic fibrosis. Here we demonstrate that a novel experimental model employing thermoneutral housing, as opposed to standard housing, resulted in lower stress-driven production of corticosterone, augmented mouse proinflammatory immune responses and markedly exacerbated high-fat diet (HFD)-induced NAFLD pathogenesis. Disease exacerbation at thermoneutrality was conserved across multiple mouse strains and was associated with augmented intestinal permeability, an altered microbiome and activation of inflammatory pathways that are associated with the disease in humans. Depletion of Gram-negative microbiota, hematopoietic cell deletion of Toll-like receptor 4 (TLR4) and inactivation of the IL-17 axis resulted in altered immune responsiveness and protection from thermoneutral-housing-driven NAFLD amplification. Finally, female mice, typically resistant to HFD-induced obesity and NAFLD, develop full disease characteristics at thermoneutrality. Thus, thermoneutral housing provides a sex- independent model of exacerbated NAFLD in mice and represents a novel approach for interrogation of the cellular and molecular mechanisms underlying disease pathogenesis.

NAFLD, a leading precursor of hepatocellular carcinoma (HCC) and recognition in combination all contribute to activation of both innate and liver transplantation1,2, encompasses a disease spectrum ranging from adaptive immune responses central to the pathogenesis of NAFLD5. benign steatosis to nonalcoholic steatohepatitis (NASH) to cirrhosis3. TLR4 polymorphisms and elevated hepatic TLR4 expression have Despite its clinical and public health importance, few effective thera- been associated with human NAFLD8,9. In addition to activating pies exist. Experimental and clinical evidence4 suggests a complex the innate immune system, TLR4 signaling also modulates multiple interplay of multiple biological processes in disease development, adaptive immune effector functions, including activation of the IL- including obesity, dysbiosis of the intestinal microbiome5,6, heightened 17 axis10. Notably, IL-17 levels correlate with obesity and NAFLD intestinal barrier permeability7, metabolic endotoxemia and various progression in mouse models11, and the transition from steatosis inflammatory processes5. Notably, HFD feeding, intestinal micro- to NASH in humans is associated with hepatic infiltration of IL-17- biome dysbiosis, augmented intestinal permeability, and metabolic producing cells12. Inactivation of the IL-17 axis inhibits progression endotoxemia and/or bacterial endotoxin (lipopolysaccharide; LPS) from steatosis to NASH in mouse models11. However, while existing © 2017 Nature America, Inc., part of Springer Nature. All rights reserved. All rights part Nature. of Springer Inc., America, Nature © 2017

1Division of Immunobiology, Department of Pediatrics, Cincinnati Children’s Hospital Research Foundation and the University of Cincinnati College of Medicine, Cincinnati, Ohio, USA. 2Immunology Graduate Program, Cincinnati Children’s Hospital Medical Center and the University of Cincinnati College of Medicine, Cincinnati, Ohio, USA. 3Department of Infectious Diseases and Microbiology, University of Lübeck, Lübeck, Germany. 4Division of Allergy and Immunology, Department of Pediatrics, Cincinnati Children’s Hospital Research Foundation and the University of Cincinnati College of Medicine, Cincinnati, Ohio, USA. 5Institute of Nutritional Medicine, University Hospital Schleswig-Holstein, Campus Lübeck, Lübeck, Germany. 6Section on Integrative Physiology and Metabolism, Joslin Diabetes Center, Boston, Massachusetts, USA. 7Institute for Diabetes and Obesity, Helmholtz Diabetes Center and German Center for Diabetes Research (DZD), Helmholtz Zentrum München, Neuherberg, Germany. 8Research Institute for Biomedical Sciences, Tokyo University of Science, Noda, Japan. 9Division of Biomedical Informatics, Department of Pediatrics, Cincinnati Children’s Hospital Research Foundation and the University of Cincinnati College of Medicine, Cincinnati, Ohio, USA. 10Division of Gastroenterology, Hepatology and Nutrition, Department of Pediatrics, Cincinnati Children’s Hospital Research Foundation and the University of Cincinnati College of Medicine, Cincinnati, Ohio, USA. 11Bill & Melinda Gates Foundation, Seattle, Washington, USA. 12Division of Pathology and Laboratory Medicine, Department of Pediatrics, Cincinnati Children’s Hospital Research Foundation and the University of Cincinnati College of Medicine, Cincinnati, Ohio, USA. 13Division of Experimental Hematology and Cancer Biology, Department of Pediatrics, Cincinnati Children’s Hospital Research Foundation and the University of Cincinnati College of Medicine, Cincinnati, Ohio, USA. 14Division of Infectious Diseases, Department of Pediatrics, Cincinnati Children’s Hospital Research Foundation and the University of Cincinnati College of Medicine, Cincinnati, Ohio, USA. Correspondence should be addressed to S.D. ([email protected]). Received 6 January; accepted 22 April; published online 12 June 2017; corrected online 21 June 2017; doi:10.1038/nm.4346

NATURE MEDICINE VOLUME 23 | NUMBER 7 | JULY 2017 829 A RTICLES

mouse NAFLD models employing both genetic (leptin deficiency) a fever after LPS challenge, whereas TN housing promotes febrile 20 and dietary (high-fat, high-carbohydrate and/or high-cholesterol responses . In the context of metabolic diseases, TN housing is required diets) interventions have proven informative, a closer recapitulation to model obesity in nude mice17, exacerbates adipose tissue inflamma- of parameters relevant to human disease is still desired. Specifically, tion24 and induces atherosclerosis in C57BL/6 wild-type (WT) mice19. mouse NAFLD models are associated with a sex bias and limited pro- Importantly, atherosclerosis, the most common cause of mortality in gression to bridging hepatic fibrosis—characteristics not observed in patients with NAFLD25, is a disease poorly modeled in WT mice. human NAFLD. These limitations, and the overall lack of representa- The relevance of TN housing to the modeling of human disease is tive animal models for preclinical testing, may be a contributing factor emerging. The majority of people in developed nations, where obesity in the paucity of therapeutic approaches for NAFLD13. is classified as a disease, tend to spend most of their day within their The temperature at which mice are typically housed in research labo- thermoneutral zone via utility of climate control inside their dwell- ratories is associated with chronic cold stress that dramatically alters ings. Further, exposure to nonthermoneutral conditions profoundly mouse physiology and immune responses14. The thermoneutral zone impacts both immune response and metabolic disease in humans. (TN), or the temperature of metabolic homeostasis, for Mus muscu- Specifically, exposure to sustained cold stress leads to dampened 15 26 lus is 30–32 °C . However, the standard temperature (TS) range at immune responsiveness to LPS challenge and improves glucose which mice are usually housed is 20–23 °C, a range chosen primarily tolerance in type 2 diabetics27. Thus, given its role in both metabo- 14 for human comfort . Housing mice under TS, as opposed to TN, con- lism and inflammation, we hypothesized that TN housing would allow ditions leads to remarkable physiological changes, including a heart for development of an improved, exacerbated and more ‘human-like’ rate increase of over 200 beats per minute, a 30% increase in mean mouse model of NAFLD. arterial blood pressure16, an overall increase in energy expenditure (50–60%)16,17 and sustained upregulation of catecholamine and corti- RESULTS 18 costeroid production . In a variety of mouse models, alleviating cold TN housing alters brown adipose tissue function and immune stress through TN housing alters immune function, including basal responsiveness cytokine production19, responses to bacterial20 and viral21,22 infection, Adaptation to cold stress involves, among other things, heightened brown 14,23 and tumor immunity . Further, mice housed at TS fail to develop adipose tissue (BAT) activity, energy expenditure and glucocorticoid

a 22 °C b c 22 °C d 22 °C A 1.5 100 * 1.5 8 30 °C * 30 °C * ** 80 22 °C * * **** * * 6 30 °C **** 1.0 1.0 B 60

value) ****

P 4 (

40 10 0.5 0.5 –log 2 20 A, C Corticosterone (ng/ml) Relative expression (AU) Relative expression (AU)

0.0 0 0.0 0 A, B, C

Nr3c1 Adrb3 Ucp1 C C Nr3c1 Adrb3 Ucp2 ° ° Ppargc1a 22 30 Ppargc1a response to infection inflammation

Decreased acute 0 1 2 –2 –1

Increased Decreasedsusceptibility inflammatory e 22 °C f g 500 * 150 3.0 2.0 * 2.0 30 °C * * * 400 2.5 100 1.5 1.5 300 2.0 * 200 50 1.5 1.0 1.0 100 * * 1.0 IL-6 (fold change) TNF (fold change) TNF (fold change) TNF (fold change) 2 4 IL-6 (fold change) 0.5 0.5 0.5 1 2

© 2017 Nature America, Inc., part of Springer Nature. All rights reserved. All rights part Nature. of Springer Inc., America, Nature © 2017 0 0 0.0 0.0 0.0 LPS –– + + LPS –– + + LPS ++++ LPS ++++ LPS + + + Spleen BMDC Spleen BMDC Cort – – +

Figure 1 Thermoneutral housing relieves stress and augments inflammation. (a–f) WT male C57BL/6 mice aged 6 weeks were housed under either standard (TS; 22 °C) or thermoneutral (TN; 30 °C) conditions for a minimum of 3 weeks. (a) Expression of the indicated genes in BAT (n = 5 per group). (b) Serum corticosterone levels (n = 7 per group). (c) Expression of the indicated genes in the spleen (n = 3 per group). AU, arbitrary units. (d) Upregulated gene expression pathways in PBMCs (left) and genes within pathways (right), determined by RNA–seq analysis (n = 2 per group). In the heat map, genes differentially regulated in known pathways include those related to ‘decreased acute inflammation’ (A), ‘increased susceptibility to infection’ (B) and ‘decreased inflammatory response’ (C). For details, see Supplementary Table 1a. (e) Fold change in serum TNF and IL-6 levels of unstimulated, TN-housed mice (n = 29); LPS-stimulated, TS-housed mice (n = 4); and TN-housed mice (n = 3) as compared to unstimulated, TS-housed mice (n = 31). (f) Fold change in splenocyte or BMDC supernatants derived from TS- or TN-housed mice for TNF (n = 3 per group) and IL-6 (n = 5 per group). (g) Serum fold change for TNF in LPS-treated, TN-housed mice (n = 3) and TN-housed mice also treated with 100 mg/l corticosterone (CORT) in drinking water (n = 4) as compared to LPS- treated, TS-housed mice (n = 4). For bar graphs, data represent mean + s.e.m. For box plots, the midline represents the mean, boxes represent the interquartile range and whiskers show the full range of values. In a–c and g, a single experiment is shown. In e, data were combined from four experiments for LPS− samples, and LPS+ samples are representative of two individual experiments. In f, a representative of two individual experiments is shown. *P < 0.05, **P < 0.01, ****P < 0.0001, Student’s t-test (a–c,e,f), hypergeometric distribution with Bonferroni correction (d) or one-way ANOVA with post hoc Tukey’s test (g).

830 VOLUME 23 | NUMBER 7 | JULY 2017 NATURE MEDICINE ARTICLES

a 30 b 40 c d e * 25 20 *** 4,000 *** 200 * *** * 20 30 15 3,000 150 20 22 °C chow *** 22 °C HFD 15 30 °C chow 20 10 2,000 100

10 30 °C HFD 10 (kcal/d) Fat mass (g) Triglycerides Weight gain (g) Lean mass (g) 10 5 1,000 50 Liver weight (mg) (mg/g liver tissue) 22 °C Chow 5 Food consumption 30 °C HFD 0 0 0 0 0 0 0 6 12 18 24 Chow HFD Chow HFD Chow HFD Chow HFD Chow HFD Time on diet (weeks)

Chow (20×) HFD (20×) HFD (40×) f 22 °C g h i j k 22 °C Chow 30 °C 30 C HFD 10 1.5 5 ° 100 Chow *** *** HFD 10 * 22 °C **** * 80 8 ** ** 4 **

(%) 8

1.0 + 6 3 * 60 6

** F4/80

4 2 + 40 0.5 4 30 °C 20 2 1 2 Oil red O (% area) CD11b NAFLD activity score Relative expression (AU) Relative expression (AU) 0 0 0 0.0 0 Srebp1c Ppara Scd Ccl2 Ccl3 Cxcl1 Chow HFD Chow HFD Chow HFD HFD HFD l m n o Col4a5 p 1,000 4 30 °C HFD Col9a2 * 6 Adamts14 800 **** Agtr2 3 *** Ngfr Accuracy of human NASH prediction 100 Srpx 600 4 (based on hepatic gene expression) 2 value) Nck2 P Osm (

Relative Ppp3cc 22 °C chow HFD 82% 10 400 10 1 * 2 * Sox4 expression (AU) Cdkn3 –log ALT (IU/l) * 30 °C chow HFD 89% CFU per 20 mg liver 200 Hk2 1 0 0 Colca2 Human healthy NASH 91% Chow HFD Acta2 Col1a1 Col1a2 Ddr1 0 Ect2 HFD Chow HFD Cdk1 Kif18b 22 °C 22 °C 30 °C 30 °C Chow Collagen formation Positive regulation of 0 1 2 HFD –2 –1

apoptotic signaling pathway

Top 200 marker genes upregulated in the ‘proliferation’ subclass of HCC

Figure 2 Thermoneutral housing exacerbates HFD-driven NAFLD pathogenesis. (a–o) WT male C57BL/6 mice aged 6 weeks were housed at TS (22 °C) or TN (30 °C) for 24 weeks and fed a chow diet or a HFD. (a) Weight gain. (b) Total body lean and fat mass. (c) Daily caloric intake. (d) Liver weight. (e) Hepatic triglyceride levels. In a–e, n = 8 per group. (f) Oil red O staining, percent area positive for lipid accumulation. (g) Representative liver histology (n = 4 per group) by H&E staining at 20× and 40× magnification. Scale bars, 100 m (left and middle) and 50 m (right). (h) NAFLD activity score. (i) Expression of the indicated lipid mediator genes in the liver. (j) Expression of the indicated chemokine genes in the liver. (k) Percent hepatic CD11b+F4/80+ immune cells. (l) Colony-forming units (CFU) of aerobic bacteria cultured from liver homogenate. (m) Expression of the indicated fibrosis marker genes in the liver. In f–m, n = 4 per group. (n) Serum ALT levels (n = 8 per group). (o) Upregulated gene expression pathways (left) and genes within pathways (right), determined by RNA–seq analysis of livers from HFD-fed mice (n = 2 per group). (p) Predictability of human NASH based on simple machine vector learning technique. Blue denotes TS (22 °C); red denotes TN (30 °C). In bar graphs, data represent mean + s.e.m. For box plots, the midline represents the mean, boxes represent the interquartile range and whiskers show the full range of values. In a–n, a representative of five individual experiments is shown; in o and p, a single experiment was performed. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, one-way ANOVA with post hoc Tukey’s test (a–f,h,k,l,n), Student’s t-test (i,j,m) or hypergeometric distribution with Bonferroni correction (o).

production28. In comparison to the mild cold stress associated with sequencing of peripheral blood mononuclear cells (PBMCs) revealed TS (22 °C), housing WT C57BL/6 mice at TN (30 °C) resulted in lower that housing at TS, in comparison to TN, resulted in greater levels of total body energy expenditure17 and expression of genes central to gene expression in pathways known to negatively regulate immune © 2017 Nature America, Inc., part of Springer Nature. All rights reserved. All rights part Nature. of Springer Inc., America, Nature © 2017 BAT activity, including glucocorticoid receptor (GR; Nr3c1), beta- responses, including pathways responsible for decreased inflamma- 3 adrenergic receptor (3AR; Adrb3; catecholamine receptor), per- tory responses, increased susceptibility to infection and decreased oxisome proliferator-activated receptor gamma coactivator 1-alpha acute inflammation (Fig. 1d and Supplementary Table 1a). (Ppargc1a) and uncoupling protein 1 (Ucp1) (Fig. 1a). Further, as We next examined the functional relevance of the altered immune compared to TS housing, TN housing resulted in lower serum concen- gene expression observed with differences in housing temperature trations of corticosterone (an immunosuppressive glucocorticoid29) for proinflammatory cytokine production. In vivo analysis revealed and splenic expression of genes encoding proteins known to inhibit that TN, in comparison to TS, housing exacerbated systemic tumor inflammatory responses, including GR, beta-2 adrenergic receptor necrosis factor (TNF) and IL-6 levels at baseline and after LPS chal- (2AR; Adrb2; catecholamine receptor), Ppargc1a and uncoupling lenge (Fig. 1e). These effects persisted ex vivo, as LPS stimulation protein 2 (Ucp2)30 (Fig. 1b,c). Notably, immune cells deficient in of splenocytes and bone-marrow-derived dendritic cells (BMDCs) GR, 2AR or UCP-2 exhibit exacerbated proinflammatory cytokine from TN-housed, as compared to TS-housed, mice resulted in height- 31–33 production following LPS stimulation . These effects were not ened TNF and IL-6 production (Fig. 1f). Changes observed at TN, unique to splenic cells, as an unbiased approach employing RNA in comparison to TS, were found to be independent of alterations

NATURE MEDICINE VOLUME 23 | NUMBER 7 | JULY 2017 831 A RTICLES

a b Week 24 c Bacteriodetes Firmicutes d 15 1.0 22 °C chow 100 Proteobacteria ) 22 °C HFD ** Firmicutes ** –4 80 0.8 30 °C chow Bacteriodetes ** 10 × 10 30 °C HFD 60 Other/ 0.6 * unclassified ** * 40 Tenericutes 0.4 5 Actinobacteria Verrucomicrobia 0.2 22 C mouse 20 Relative abundance ° FITC-dextran (

Phylum composition (%) 30 °C mouse 0 0.0 0 NASH-like metagenome Normal human 22 °C 30 °C 22 °C 30 °C C C C C 30 60 90 120 150 180 ° ° ° ° NASH human 22 30 22 30 Time (min) Chow HFD NormalNASH NormalNASH PC2 Mouse Human Mouse Human PC1 e 22 °C Ctrl (chow) f 22 °C (40×) 30 °C (40×) g h 22 °C Ctrl (HFD) 15 30 °C Ctrl (HFD) 4 1,500 ** *** ) 22 °C Tx (HFD) *** –4 30 °C Tx (HFD) * 22 °C 10 Ctrl 3 × 10 30 °C *** 1,000 Untreated 2 Tx 5 ALT (IU/l) 500 1 FITC-dextran ( Inflammation score

0 0 0 30 60 90 120 150 180 Abx Tx HFD: Ctrl Tx HFD: Ctrl Tx Time (min)

Figure 3 Thermoneutral housing is associated with augmented intestinal permeability and dysbiosis of the microbiome. (a–c) WT male C57BL/6 mice aged 6 weeks were housed at TS (22 °C) or TN (30 °C) for 24 weeks and fed a chow diet or a HFD. (a) Flow of FITC-labeled dextran through the intestine (n = 4 per group). (b) Phylum-level differences between TS- and TN-housed mice after 24 weeks of HFD (n = 8 per group). (c,d) Relative phylum abundance (c) and principal-coordinate analysis (d) in mice housed at TS (n = 8 per group) or TN (n = 8 per group) and either healthy, lean humans 6 (n = 16 per group) or patients with NASH (n = 22 per group) described previously . (e–h) WT male C57BL/6 mice aged 6 weeks were housed at TS (22 °C) or TN (30 °C) and fed a HFD. After 8 weeks of HFD, mice were mock treated (Ctrl) or treated with antibiotics (Tx) supplied in drinking water in addition to HFD exposure for an additional 16 weeks. (e) Flow of FITC-labeled dextran through the intestine of control chow- and HFD-fed mice at 22 °C and HFD-fed mice at 30 °C (n = 4 per group) and treated HFD-fed mice at 22 °C and 30 °C (n = 3 per group). (f) Representative liver histology (n = 4 per group) by H&E staining. Scale bars, 50 m. (g) Inflammation score per NAFLD activity score criteria (n = 4 per group). (h) Serum ALT levels (n = 8 per group). For bar graphs, data represent mean + s.e.m. For box plots, the midline represents the mean, boxes represent the interquartile range and whiskers show the full range of values. In a, a representative of two individual experiments is shown; in b–h, a single experiment was performed. *P < 0.05, **P < 0.01, ***P < 0.001, one-way ANOVA with post hoc Tukey’s test (a,e,g,h) or Mann–Whitney test (c).

in cellularity and composition of bone marrow, PBMCs, spleen and dietary exposure nullified differences in weight gain and body thymus (Supplementary Fig. 1a–d). These findings indicate that TN lean or fat mass (Fig. 2a,b). Similar weight gain occurred despite housing reverses the inhibition of immune responsiveness seen under lower food intake at TN (Fig. 2c), likely owing to lower energy standard housing conditions. expenditure at TN (refs. 16,17). However, despite similar body The anti-inflammatory effects of glucocorticoids and corticosterone weights and adiposity, at the time of harvest, TN-housed mice dis- are well established. We thus examined the relevance of the corticos- played heightened visceral adipose tissue immune cell infiltration and terone axis to immune responsiveness at TN. Administration of corti- adipose tissue macrophage activation in comparison to TS-housed, costerone to TN-housed mice was sufficient to reverse the exacerbated HFD-fed mice (Supplementary Fig. 2a–c)—in agreement with a 24 proinflammatory cytokine production associated with TN, in com- previous report . parison to TS, housing (Fig. 1g), suggesting that the corticosterone Given the role of obesity and adipose tissue inflammation in glu- 35 axis is functional at TN and plays a role in the regulation of inflam- cose dysmetabolism , we next evaluated the impact of TN housing on matory vigor. Conserved effects were observed in humans following glucose homeostasis. T , in comparison to T , housing exacerbated

© 2017 Nature America, Inc., part of Springer Nature. All rights reserved. All rights part Nature. of Springer Inc., America, Nature © 2017 N S stimulation of human PBMCs with LPS in the presence of GR and glucose intolerance only when differences in body weight existed 34 2AR agonists (data not shown) . In sum, these results suggest that (Supplementary Fig. 2d). These differences in glucose tolerance were ambient temperature profoundly alters BAT function, glucocorticoid diminished once body weights were normalized (Supplementary production and host innate immune responses in male C57BL/6 mice Fig. 2e). Further, with similar weight gains, TN- and TS-housed and are congruent with reports indicating that TN housing promotes mice displayed similar insulin sensitivities (Supplementary Fig. 2f), a more human-like immune response in mice20. which correlated with similar islet sizes and hepatic levels of AKT phosphorylation (data not shown). These data agree with previous TN housing exacerbates HFD-driven NAFLD pathogenesis reports describing the role of TN housing in the modulation of glucose As the immune system plays a central role in the pathogenesis of obesity- metabolism and insulin resistance in obesity24. associated sequelae, we hypothesized that TN, in comparison to TS, We next examined the impact of TN housing on NAFLD develop- housing would exacerbate such effects in male WT C57BL/6 mice. TN ment and progression. Despite having similar serum triglyceride levels housing in combination with HFD feeding initially accelerated weight and serum and hepatic cholesterol levels (Supplementary Fig. 2g–i), gain in comparison to TS-housed HFD-fed mice; however, prolonged obese mice housed at TN, in comparison to TS, had exacerbated liver

832 VOLUME 23 | NUMBER 7 | JULY 2017 NATURE MEDICINE ARTICLES

a Il5 b 22 °C (40×) 30 °C (40×) c d Cytokines and Il10 10 1,000 * inflammatory response ** Il15 22 °C fl/fl 800 Positive regulation Lbp Tlr4 ; 8 30 °C Tlr5 Vav1-Cre– fl/fl – of TLR signaling *** 6 600 Tlr4 ; Vav1-Cre Ly96 fl/fl + Cellular response Nr1d1 Tlr4 ; Vav1-Cre * 30 °C 4 400

to LPS ALT (IU/l) HFD Cyp17a1 Ltf 200 Positive regulation fl/fl 2 of TLR4 signaling * 22 °C 30 °C Tlr4 ; Vav1-Cre+ NAFLD activity score 0 0 fl/fl 0 1234 fl/fl Tlr4 – – ++ 0 1 2 Tlr4 ––++ –2 –1 Vav1-Cre Vav1-Cre –log10 (P value) HFD HFD

Figure 4 Thermoneutral-housing-driven modulation of hematopoietic TLR4 signaling regulates NAFLD progression. (a) WT male C57BL/6 mice aged 6 weeks were housed under TS (22 °C) or TN (30 °C) conditions for 2 weeks before initiation of 8 weeks of HFD. Upregulated gene expression pathways (left) and genes within pathways (right) are shown, as determined by RNA–seq analysis (n = 2 per group). (b–d) Tlr4fl/fl; Vav1-Cre+ and control Tlr4fl/fl; − Vav1-Cre mice on a C57BL/6 background were housed at TS (22 °C) or TN (30 °C) for 24 weeks and fed a chow diet or a HFD. (b) Representative liver histology by H&E staining at 40× magnification of Tlr4fl/fl; Vav1-Cre− mice housed at 22 °C (n = 7 per group) and 30 °C (n = 5 per group) as compared to Tlr4fl/fl; Vav1-Cre+ mice housed at 22 °C (n = 4 per group) and 30 °C (n = 4 per group). Scale bars, 50 m. (c) NAFLD activity score as determined by histology. (d) Serum ALT levels of Tlr4fl/fl; Vav1-Cre− mice housed at 22 °C (n = 9 per group) and 30 °C (n = 8 per group) as compared to Tlr4fl/fl; + Vav1-Cre mice housed at 22 °C (n = 7 per group) and 30 °C (n = 5 per group). Blue denotes TS (22 °C); red denotes TN (30 °C). For box plots, the midline represents the mean, boxes represent the interquartile range and whiskers show the full range of values. In a, a single experiment is shown; in c and d, a representative of two independent experiments is shown. *P < 0.05, **P < 0.01, ***P < 0.001, hypergeometric distribution with Bonferroni correction (a) or one-way ANOVA with post hoc Tukey’s test (c,d).

39 weight and hepatic steatosis, as quantified by hepatic triglyceride in humans revealed that TN housing resulted in similar changes in levels, oil red O staining and NAFLD activity score36 (Fig. 2d–h). the expression of genes associated with a variety of NAFLD-related Histological analysis, used to determine the NAFLD activity score, pathways (e.g., inflammatory response, response to reactive oxygen suggested that hepatocytes from TN-housed, in comparison to species and leukocyte activation; Supplementary Fig. 3g). TS-housed, obese mice exhibited elevated steatosis and hepatocyte Next, we determined whether changes in HFD-induced hepatic ballooning, characterized by cellular swelling, rarefaction of the gene expression at TN or TS were more likely to predict the gene hepatocytic cytoplasm and clumped strands of intermediate fila- expression patterns induced by human NASH39. Use of support vector ments (Fig. 2g). These findings also correlated with reduced gene machine analysis40 determined that the gene expression differences expression of key lipid mediators known to have decreased expression induced by a HFD at TN allowed for improved prediction of the gene 37,38 during NASH (Fig. 2i) . expression patterns of human NASH in comparison to TS housing We next examined the progression to and severity of NASH under and a HFD (Fig. 2p). These findings verify markedly exacerbated TN housing. Modest changes in the expression of genes related to lipid HFD-driven NAFLD pathogenesis and hepatocellular damage at TN handling, chemokine production and fibrosis were observed in mice and suggest an improved model for human disease. fed a chow diet (Supplementary Fig. 3a–c). In the context of HFD Although most commonly used for obesity and NAFLD mode- feeding, however, TN-housed mice exhibited robust exacerbation of ling, the C57BL/6 mouse strain is highly resistant to the induction hepatic chemokine expression (Fig. 2j)3, macrophage infiltration of of hepatic fibrosis. Thus, we next determined whether the lack of the liver (Fig. 2k) and bacterial translocation to the liver (Fig. 2l), overt hepatic fibrosis observed in C57BL/6 mice was strain specific. in comparison to TS-housed mice. Further, TN-housed, in compari- Notably, the AKR mouse strain develops robust obesity and NAFLD 41 son to TS-housed, obese mice also displayed elevated expression of when fed a HFD . When housed at TN, as opposed to TS, and fed a genes associated with induction of hepatic fibrosis (Fig. 2m) and a HFD, AKR mice gained more weight but maintained similar visceral threefold induction in hepatocellular damage, as measured by serum and subcutaneous adiposity (Supplementary Fig. 4a,b). TN-housed, alanine transaminase (ALT) levels (Fig. 2n). However, despite these obese AKR mice exhibited greater liver weight, hepatic steatosis and robust changes, the induction of overt bridging hepatic fibrosis was NAFLD activity scores than TS-housed AKR controls (Supplementary not observed in WT C57BL/6 mice (Supplementary Fig. 3d). Fig. 4c–f). The lower NAFLD activity scores observed in AKR mice as To begin to examine the effect of T N housing on hepatic gene expres- compared to C57BL/6 mice may be mouse strain dependent and/or due sion in the presence or absence of dietary modulation, we performed to shortened HFD exposure. The NAFLD activity score in AKR mice © 2017 Nature America, Inc., part of Springer Nature. All rights reserved. All rights part Nature. of Springer Inc., America, Nature © 2017 an unbiased approach using RNA–seq analysis of liver tissue. Although was associated with greater lobular inflammation and steatosis at TN, TN, in comparison to TS, housing altered hepatic gene expression but limited hepatocyte ballooning was observed in both groups at this in chow-fed mice, a HFD exacerbated differential gene expression time. Augmented lobular inflammation correlated with higher hepatic (Supplementary Fig. 3e). In chow-fed mice, analysis of differential chemokine gene expression, macrophage infiltration and hepatocellu- gene expression revealed alterations in liver metabolism, including lar damage (Supplementary Fig. 4g–i) in TN-housed, as compared to elevated expression of gene pathways associated with lipid metabolism TS-housed, AKR mice. Notably, unlike in C57BL/6 mice, HFD feeding and fatty acid oxidation (Supplementary Fig. 3e,f and Supplementary under conditions of TN but not TS housing was sufficient to induce Table 1b). In contrast, analysis of differential gene expression in HFD- fibrosis in AKR mice, as quantified by hepatic gene expression and fed mice showed heightened expression of genes and gene pathways Trichrome staining (Supplementary Fig. 4j,k). Together, these data associated with collagen formation, apoptosis and HCC induction indicate that the effects of TN housing on NAFLD pathogenesis are (Fig. 2o and Supplementary Table 1c). Further, comparison of the conserved across different mouse strains and suggest that the use of hepatic gene expression changes induced by TN versus TS housing in obese AKR mice at TN may provide a novel and accelerated model for HFD-fed mice to known gene expression changes induced by NASH mechanistic interrogation of NAFLD-induced hepatic fibrosis.

NATURE MEDICINE VOLUME 23 | NUMBER 7 | JULY 2017 833 A RTICLES

a bc22 °C 22 °C dWT Il17ra–/– Il17a–/– ef 30 °C 30 °C 9 1,000 8 * 30 C 2.5 ** ° (%) ** 30 °C 8 800 Increased IL-17 ** + WT (%) 6 2.0 (20×) secretion ** + 7 600 Il17ra–/– TNF 1.5 –/– Il1rn 4 + Il17a Nfil3 6 400 IL-17A 0 123 1.0 ALT (IU/l) Rag1 + 5 200 Clec6a –log (P value) 2 30 °C 10 IL-17A 0.5 Il17rc +

CD4 (40 ) × NAFLD activity score 4 0 22 °C 30 °C 30 °C HFD 0 0.0 –/– –/– –/– –/– HFD CD4 HFD WT WT 0 1 2 –2 –1 Il17ra Il17a Il17ra Il17a

HFD HFD

Figure 5 Thermoneutral housing is associated with pathogenic upregulation of the IL-17 axis. (a) WT male C57BL/6 mice aged 6 weeks were housed under TS (22 °C) or TN (30 °C) conditions for 2 weeks before initiation of 8 weeks of HFD feeding. The upregulated gene expression pathway (top) and genes within the pathway (bottom) are shown, as determined by RNA–seq analysis (n = 2 per group). (b,c) WT male C57BL/6 mice aged 6 weeks + + + were housed at TS (22 °C; n = 3 per group) or TN (30 °C; n = 4 per group) for 24 weeks and fed a HFD. (b) Percentage of TCR- CD4 IL-17A hepatic infiltrating immune cells. (c) Percentage of TCR-+CD4+IL-17A+TNF+ hepatic infiltrating immune cells. (d–f) WT (n = 5 per group), Il17ra−/− (n = 4 per −/− group) and Il17a (n = 6 per group) male mice aged 6 weeks, all on the C57BL/6 background, were housed under TN (30 °C) conditions for 2 weeks before initiation of 24 weeks of HFD feeding. (d) Representative liver histology of WT (n = 5 per group), Il17ra−/− (n = 4 per group) and Il17a−/− (n = 6 per group) mice, by H&E staining at 20× and 40× magnification. Scale bars, 100 m (top) and 50 m (bottom). (e) NAFLD activity score. (f) Serum ALT levels. For box plots, the midline represents the mean, boxes represent the interquartile range and whiskers show the full range of values. In a, a single experiment is shown; in b–f, a representative of three individual experiments is shown. *P < 0.05, **P < 0.01, hypergeometric distribution with Bonferroni correction (a), Student’s t-test (b,c) or one-way ANOVA with post hoc Tukey’s test (e,f).

The adrenal glands are a central production site for circulating Although the microbiome plays a role in NAFLD in both humans corticosterone and catecholamines. Immune responsiveness under and mice, the two species display an inherently different microbial cold stress is, in part, modulated by corticosterone and catecholamine composition. Nevertheless, we examined whether the mouse micro- levels. Adrenalectomized male C57BL/6 mice fed a HFD exhibited biome at TN correlated more closely with microbiomes reported in similar weight gains and visceral and subcutaneous adiposity at humans with NASH6. Comparison of 16S rRNA sequences demon- the two housing temperatures (Supplementary Fig. 5a–d). Unlike strated that TN housing of mice on a HFD led to greater similarity of C57BL/6 mice with intact adrenal glands, adrenalectomized C57BL/6 the gut microbiome to that observed in people with NASH6. Notably, mice housed at TN and TS exhibited similar liver weights, hepatic stea- this was seen in both phylum-level alterations (Fig. 3c) and principal- tosis, lipid mediators and chemokine gene expression, macrophage component analysis (Fig. 3d), where an upward shift (PC2) corre- infiltration, expression of genes associated with induction of fibrosis sponded with a more NASH-like metagenome. and hepatocellular damage (Supplementary Fig. 5e–l). Notably, the TN housing in combination with stress from a HFD promoted lack of adrenal glands in mice with TS housing in combination with a expansion of Gram-negative Bacteriodetes (Fig. 3b). Antibiotic- HFD exacerbated hepatocellular damage as compared to nonadrenal- mediated depletion of the Gram-negative microbiome in HFD-fed ectomized mice at TS and resulted in levels of damage similar to those mice housed at TS and TN did not alter total body weight gain, vis- observed in TN-housed mice (Supplementary Fig. 5m). These data ceral and subcutaneous adiposity, hepatic weight or hepatic triglyc- suggest that mediators produced by the adrenal glands play an impor- eride accumulation, although TS-housed, antibiotic-treated mice tant role in suppressing NAFLD progression during cold stress. displayed elevated glucose intolerance (Supplementary Fig. 8a–g). However, antibiotic treatment obviated the greater intestinal per- Intestinal permeability and microbiome in TN-driven NAFLD meability observed in HFD-fed mice housed at TN, as compared 7 Augmented intestinal permeability and dysbiosis of the intestinal to mice housed at TS (Fig. 3e), and lowered the NAFLD activity microbiome6 contribute to human and mouse NAFLD progression4–6. inflammation score (Fig. 3f,g) and protected against hepatocel- Although histological analysis of the small intestine did not reveal dif- lular damage (Fig. 3h) only at TN. Of note, these protective effects ferences due to housing temperature or diet in immune cell infiltration in NAFLD were specific to obese TN-housed mice and were not (Supplementary Fig. 6a), TN, as compared to TS, housing exacerbated observed in obese, antibiotic-treated TS-housed mice (Fig. 3f–h and paracellular intestinal permeability, as evidenced by translocation of Supplementary Fig. 8g). These data indicate that changes in intestinal FITC-dextran across the epithelium, and lowered transepithelial microbiome composition are associated with T -driven amplification

© 2017 Nature America, Inc., part of Springer Nature. All rights reserved. All rights part Nature. of Springer Inc., America, Nature © 2017 N resistance on both a chow diet and a HFD (Fig. 3a and Supplementary of NAFLD. Fig. 6b). Further, TN, as compared to TS, housing changed the intestinal microbiome in mice in as little as 2 weeks, before the mice were Hematopoietic TLR4 and IL-17 axis in TN-driven NAFLD placed on a HFD (Supplementary Fig. 6c,d). Extended exposure LPS from Gram-negative bacteria activates TLR4. Our data indicate (12 weeks) of randomly separated WT C57BL/6 mice to different that TLR4 responsiveness is elevated at thermoneutrality (Fig. 1e,f). ambient temperatures or a HFD further exacerbated these changes, Additionally, TLR4 polymorphisms and hepatic TLR4 expression have with obvious differences observed in every phylum analyzed after 24 been correlated with the progression of human NAFLD9. Whether weeks of exposure (Fig. 3b and Supplementary Fig. 6e–k). Notably, TN-driven modulation of innate immune responses is maintained on TN housing enriched the representation of Bacteroidetes, whereas TS a HFD has not been examined. RNA–seq analysis of PBMCs from WT housing preferentially enriched for Firmicutes. Linear discriminant C57BL/6 mice challenged with a HFD revealed greater levels of gene analysis effect size (LEfSe) analysis, which was employed to analyze expression in pathways associated with cytokine production and TLR data with specificity to the genus level, revealed congruent differences responsiveness (Fig. 4a and Supplementary Table 1d). To determine beyond the phylum level (Supplementary Fig. 7). the functional relevance of differential TLR4 expression in immune

834 VOLUME 23 | NUMBER 7 | JULY 2017 NATURE MEDICINE ARTICLES

22 °C chow 30 °C chow 22 C Chow Chow (20) HFD (20) HFD (40) a 22 °C HFD 30 °C HFD b c d e 30 C HFD 35 **** 40 25 * 2,000 * 250 30 **** ***

**** 200 25 30 20 1,500 22 C 20 15 150 15 20 1,000 10 *** 10 100

Fat mass (g) 10 Weight gain (g) 5 500 Triglycerides Lean mass (g) 5 50 30 C Liver weight (mg) 0 (mg/g liver tissue) 0 0 0 0 06 12 18 24 Chow HFD Chow HFD Chow HFD Chow HFD Time on diet (weeks)

22 C 22 C 22 C 22 C 22 C f g 30 C h 30 C i 30 C j k *** 6 30 C 25 10 8 2.0 30 C 800 **** ** ** ** 22 °C ** (%) *

20 8 +

(%) 6 1.5 600 30 C (%) ° – 4 ** +

15 6 TNF Chow +

Gr-1 4 1.0 400 HFD +

10 IL-17A 4 Relative Relative +

2 ALT (IU/l)

IL-17A 2 0.5 200 5 2 + expression (AU) expression (AU) CD4 CD11b

0 0 0 CD4 0 0.0 0 Ccl2 Ccl3 Cxcl1 HFD HFD HFD Acta2 Col1a1 Col1a2 Chow HFD HFD HFD l Brown adipocyte UCP-1 Energy expenditure 3AR

Intestine HFD NAFL Microbiome

Decreased 30 C norepinephrine housing corticosterone Gram–negative skewing Permeability/ LPS

TLR4 signaling TNF, IL-6, IL-1, IL-23 NASH cAMP IL-17 2AR /GR UCP-2

APC Th17

Figure 6 Thermoneutral housing removes the barrier to modeling obesity and NAFLD in female mice. (a–k) WT female C57BL/6 mice aged 6 weeks were housed at TS (22 °C) and fed chow (n = 4 per group) or a HFD (n = 6 per group) or housed at TN (30 °C) and fed chow (n = 4 per group) or a HFD (n = 6 per group) for 24 weeks. (a) Weight gain. (b) Total body fat mass and lean mass. (c) Liver weight. (d) Hepatic triglyceride levels. (e) Representative liver histology (n = 4 per group) by H&E staining at 20× and 40× magnification. Scale bars, 100 m (left and middle) and 50 m (right). (f) Expression of the indicated chemokine genes in the liver of HFD-fed mice. (g) Percentage of hepatic infiltrating CD11b+GR-1− immune cells in HFD-fed mice. (h) Percentage of hepatic infiltrating TH17 cells in the total hepatic immune cell infiltrate in HFD-fed mice. (i) Percentage of hepatic infiltrating TH17 cells also producing TNF in the total hepatic immune cell infiltrate in HFD-fed mice. (j) Expression of the indicated fibrosis marker genes in the liver in HFD-fed mice. (k) Serum ALT levels in chow- and HFD-fed mice. (l) Schematic depicting the role of TN housing in promoting exacerbated NAFLD pathogenesis during HFD feeding where red lines represent factors associated with exacerbated NAFLD at TN (APC, antigen-presenting cells; NAFL, nonalcoholic fatty liver disease). This schematic, in part, is composed of an available artwork (http://www.servier.com/ Powerpoint-image-bank), reproduced per a licensing agreement (https://creativecommons.org/licenses/by/3.0/legalcode). For bar graphs, data represent mean + s.e.m. For box plots, the midline represents the mean, boxes represent the interquartile range and whiskers show the full range of values.

© 2017 Nature America, Inc., part of Springer Nature. All rights reserved. All rights part Nature. of Springer Inc., America, Nature © 2017 In a–k, a representative of three individual experiments is shown. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, one-way ANOVA with post hoc Tukey’s test (a–d,k) or Student’s t-test (f–j).

fl/fl 42 cells, Tlr4 mice were used. Congruent with a previous report , removing suppression of TLR4 signaling through TN housing can Vav1-Cre-driven hematopoietic cell deletion of Tlr4 in TS-housed mice contribute to NAFLD pathogenesis. did not hinder the progression of HFD-induced NAFLD (Fig. 4b–d). Induction of IL-6, IL-1 and IL-23 production is associated with 10 However, under TN housing, such deletion was sufficient to protect T helper type 17 (TH17) cell polarization and IL-17 axis activation . from HFD-driven increases in weight gain, visceral and subcutane- Hepatic IL-6 expression is elevated in human NAFLD, and serum ous adiposity, glucose intolerance, liver weight, histologically iden- concentrations of IL-1 are higher in individuals with metabolic syn- tifiable steatosis, hepatocyte ballooning, and lobular inflammation drome43,44. RNA–seq analysis of PBMCs from WT C57BL/6 mice pro- and hepatocellular damage (Fig. 4b–d and Supplementary Fig. 9). vided initial suggestions that TN, as compared to TS, housing coupled These data indicate that TLR4 signaling is fundamentally modu- with a HFD resulted in greater expression of genes related to IL-17 lated by ambient temperature in the context of HFD feeding and that production (Fig. 5a and Supplementary Table 1d). We next examined

NATURE MEDICINE VOLUME 23 | NUMBER 7 | JULY 2017 835 A RTICLES

whether TN housing altered TLR4-signaling-driven induction of with experimental modeling of mouse obesity and NAFLD. In sum, mediators associated with activation of the IL-17 axis. LPS stimula- these findings demonstrate that TN-housing-driven NAFLD patho- tion of BMDCs from TN-housed, as compared to TS-housed, obese genesis represents a novel disease model associated with altered cor- mice resulted in greater IL-6 and IL-1 production and greater Il23a ticosterone levels, immune responses, metabolism, intestinal barrier gene expression (Supplementary Fig. 10a–c). Elevated production integrity and intestinal microbiome (Fig. 6l). of these mediators correlated with exacerbated hepatic infiltration of CD4+ T cells capable of both single IL-17 and dual IL-17 and TNF DISCUSSION production in TN-housed, obese mice (Fig. 5b,c). Notably, dual IL-17- NAFLD research often involves the use of mouse models housed at a and TNF-producing CD4+ T cells exacerbate pathogenesis in mouse suboptimal, cold-stress-inducing ambient temperature (typically 22 °C) experimental autoimmune encephalomyelitis models45 and have been that fails to comprehensively recapitulate human disease. In the associated with higher severity of Crohn’s disease in humans45,46. case of NAFLD, current models in mice include a fundamental sex Hence, we examined the contribution of IL-17 axis activation to TN- bias, unlike in humans; an inability to study the interplay between driven NAFLD pathogenesis. Despite similar weight gains, visceral atherosclerosis and NAFLD; limited induction of pathways associ- and subcutaneous adiposity, hepatic steatosis and NAFLD activity ated with the development of hepatic fibrosis and HCC; and altered scores (Fig. 5d,e and Supplementary Fig. 10d–h), TN-housed, obese immune responsiveness as compared to humans with this condition. IL-17 axis–deficient mice (Il17ra−/− and Il17a−/−) were protected Although previous reports have demonstrated that housing mice from glucose intolerance, exacerbated liver weight and hepatocel- at a thermoneutral temperature (30 °C) affects the pathogenesis of lular damage, as compared to TN-housed WT controls (Fig. 5f and multiple infectious and metabolic complications, whether TN hous- Supplementary Fig. 10i,j). These findings suggest that the IL-17 axis ing allows for an improved mouse NAFLD model has not previously is a relevant factor in the regulation of NAFLD pathogenesis and that, been interrogated. unlike TLR4 signaling, a role for the IL-17 axis in NAFLD is conserved Here we demonstrate, in accordance with previous reports, that the across housing temperatures. cold stress associated with standard mouse housing procedures inhib- 14 Serum amyloid A (SAA), an acute-phase protein largely produced its immune responses . Housing mice at TN upregulated immune in the liver, is known to both activate TLR4 and promote TH17 cell responsiveness both in vivo and ex vivo, and this upregulation was 47 differentiation via antigen-presenting cells . TN-housed, obese mice suppressed by exogenous administration of corticosterone. Further, exhibited greater hepatic SAA1 and SAA2 expression in comparison TN housing exacerbates cellular responsiveness to inflammatory lig- to TS-housed mice (Supplementary Fig. 11a), whereas deletion of ands, without alteration of the cell type composition or cell numbers. Il17ra, Il17a or hematopoietic Tlr4 was associated with lower hepatic Additional studies, however, are required to functionally examine expression of SAA1 (Supplementary Fig. 11b,c). These data suggest the pathways central to altered immune cell responsiveness at TN. that TN-mediated modulation of SAA production might represent a As such changes are sustained ex vivo, TN-driven modulation of cel- link between activation of the TLR4 and IL-17 axes. lular metabolism and epigenetics are likely to play a role and warrant further investigation. TN housing allows for disease modeling in female mice Robust exacerbation of NAFLD pathogenesis in the context of 48 Human NAFLD prevalence is similar in men and women . However, TN housing and HFD-induced stress correlated with heightened hepatic protection from severe HFD-induced obesity and NAFLD in C57BL/6 steatosis, hepatic immune cell infiltration, elevated expression of genes female mice precludes the interrogation of disease pathogenesis in a associated with hepatic fibrosis and HCC, and hepatocellular damage. sex-independent manner. Hence, we asked whether TN housing would Therefore, TN-driven amplification of disease-propagating processes allow for the induction of obesity and NAFLD in female C57BL/6 may allow for in-depth interrogation and improved definition of the mice. As expected, TS-housed female mice fed a HFD, as compared mechanisms central to NAFLD pathogenesis, including critical sites of to chow, displayed mild weight gain over time (Fig. 6a). However, TN- inflammation, cell type(s) and immune mediators, using a genetically housed female mice fed a HFD, as compared to chow, exhibited robust unbiased approach. However, despite robust disease exacerbation, obesity and total body adiposity (Fig. 6a,b), with these increased to hepatic fibrosis was not observed in male or female WT C57BL/6 the levels observed in TN-housed male mice (Fig. 2). Further, in agree- mice. Thus, additional examination of hepatic fibrosis development ment with data from male mice, under conditions of robust weight in C57BL/6 mice, using TN housing in combination with dietary gain, obese, TN-housed female mice displayed exacerbated glucose challenges known to promote fibrosis (for example, methionine- intolerance (Supplementary Fig. 12a). choline-deficient diets, high-fat plus high-cholesterol diets and Similarly, T -housed, obese female mice exhibited greater liver high-fat plus high-fructose diets)49, might offer a clear advantage over

© 2017 Nature America, Inc., part of Springer Nature. All rights reserved. All rights part Nature. of Springer Inc., America, Nature © 2017 N weight, hepatic steatosis, hepatic chemokine expression3, macrophage existing experimental NAFLD models for evaluating novel therapeu- infiltration and hepatic infiltration of both IL-17-producing and dual tics. Notably, the induction of measureable hepatic fibrosis by a HFD IL-17- and TNF-producing TH17 cells, as compared to TS-housed, at TN is possible, as demonstrated in AKR mice. Employing these HFD-fed mice (Fig. 6c–i). Further, as in WT male C57BL/6 mice, mice in future studies may provide a novel model for mechanistic these physiological changes in the liver of obese TN-housed, as com- interrogation of NAFLD-induced hepatic fibrosis. Lastly, TN housing pared to TS-housed, female mice correlated with greater expression appears to hold relevance to human disease and improve upon exist- of fibrosis-associated genes (Acta2, Col1a1 and Col1a2) and a greater ing NAFLD models, even in C57BL/6 mice, as hepatic gene expression degree of hepatocellular damage (Fig. 6j,k) but did not induce overt in this model provides improved prediction of human NASH. bridging hepatic fibrosis (Supplementary Fig. 12b). Notably, TN- Interactions between the microbiome and the immune system are housed, as compared to TS-housed, obese WT female mice exhibited thought to play a key role in NAFLD. Augmented intestinal perme- 7 greater Saa1 expression (Supplementary Fig. 12c). Thus, TN housing ability also correlates with NASH severity . Notably, TN housing and allows for development of robust obesity and NAFLD in WT female a HFD resulted in heightened intestinal permeability and intestinal C57BL/6 mice and, notably, removes the sex bias typically associated microbiome dysbiosis, which mirrors the obesity- and NASH-driven

836 VOLUME 23 | NUMBER 7 | JULY 2017 NATURE MEDICINE ARTICLES

6 dysbiosis observed in humans . Intestinal microbiome changes are relevance of TN housing to human physiology and allows for the reali- certainly one of the many factors in the causal nexus that allow pro- zation of novel pathways associated with protection from HFD-driven nounced expression of the NAFLD phenotype in mice housed at obesity in female mice. Notably, TN housing may allow for improved thermoneutrality. However, given the consistency in the impact of modeling of the contribution of maternal obesity to disease patho- TN housing on the modulation of immune responses and metabo- genesis over generations of offspring. In addition, atherosclerosis, lism among multiple research institutions, it is unlikely that a unique the most common cause of mortality in patients with NAFLD25, is ‘microbiome mix’, specific to a single institution, is driving observed another metabolic disease that lacks an appropriate mouse model as differences independently of temperature effects. Notably, our data current models are similarly sex restrained. Housing WT mice at TN in evoke several interesting questions and areas of future study: what combination with a ‘Western’ diet allows for atherosclerosis induction are the bidirectional interactions between altered microbiome and in male C57BL/6 mice19 and, hence, might provide for interrogation altered immune responsiveness under TN housing? Does TN housing of the pathways critical for the interplay of these two diseases, as well alter the immune response in gnotobiotic mice? Can protection from as the removal of sex bias in modeling atherosclerosis. 50 HFD-driven obesity in germ-free mice be reversed using TN hous- Thus, housing temperature is an overlooked and very important ing? Are levels of specific microbial byproducts or microbial species variable in the modeling of human disease. Close attention to this thought to play a role in obesity-associated inflammation and NAFLD variable has promise for the discovery of novel mechanisms underly- altered at TN (refs. 11,51)? ing disease pathogenesis and for improved modeling of the noncom- Our data suggest that TLR4 signaling, presumably via Gram-negative municable metabolic diseases that are causing an increasing burden microbiome dysbiosis, plays a role in TN-driven NAFLD patho- of morbidity and mortality worldwide. genesis. Additionally, we demonstrate that this pathogenic role for Vav1-driven deletion of TLR4 expression is important only when mice METHODS are housed at TN. Of note, Vav1-Cre can be activated in different Methods, including statements of data availability and any associated cells, and the necessary hematopoietic and/or endothelial cell type accession codes and references, are available in the online version of should be evaluated52. Also, whether the role of TLR4 is dependent the paper. on LPS or induction of various endogenous TLR4 ligands (e.g., high- Note: Any Supplementary Information and Source Data files are available in the mobility-group box 1 protein (HMGB1), fibronectin, fibrinogen and online version of the paper. resistin) that have been associated with NAFLD progression42,53–55 is unknown. ACKNOWLEDGMENTS Our findings also highlight the functional effector relevance of the This work was supported in part by NIH R01DK099222 and R01DK099222-S1 (to S.D.), the CCHMC Pediatric Diabetes and Obesity Center initiative (to S.D.), IL-17 axis in TN-housing-driven NAFLD. The cellular and molecular R01DK033201 (to C.R.K.), K12-HD000850 (to S.S.), NIH T32AI118697 (associated mechanisms underlying the IL-17-mediated effects in NAFLD are with D.A.G.), NIEHS Grant P30 ES006096 University of Cincinnati Center for not well defined. IL-17-driven induction of chemokines (e.g., chem- Environmental Genomics (associated with D.A.G.), PHS Grant P30 DK078392 okine (C-X-C motif) ligand 1 (CXCL1) and C-C motif chemokine Pathology of the Digestive Disease Research Core Center at CCHMC (associated with ligand 2 (CCL2)) is associated with both macrophage and neutrophil S.D.) and German Research Foundation IRTG 1911 (projects A6 and B8 to C.S. and J.R.). We would also like to acknowledge C. Chougnet (CCHMC) for providing access 56 hepatic infiltration and activation . Further, it has been demon- to human PBMC samples and C. Woods (CCHMC) for technical assistance. strated that IL-17 plays an important role in liver fibrosis57. Notably, hepatic expression of both Cxcl1 and Ccl2 and markers of fibrosis AUTHOR CONTRIBUTIONS is elevated in livers from T -housed, HFD-fed mice. Thus, defini- D.A.G., M.E.M.-F., T.E.S., S.G., M.C., D.W., R.M., C.C.C., M.J.L., J.K., S.S., A.S. and N D.R. participated in data generation. D.A.G., M.E.M.-F., S.G., D.R., R.K., B.J.A., tion of the critical IL-17RA-expressing cell type(s) within the liver S.K.S., R.S., K.A.S., D.B.H., J.R., S.P.H. and S.D. participated in data analysis and (e.g., hepatocytes, stellate cells, Kupffer cells, etc.) deserves examina- interpretation. S.S., K.S., Y.I., C.R.K., B.J.A., C.S. and C.L.K. provided materials and tion. Similarly, definition of the critical IL-17-producing cell type(s) technical support and participated in critical review of the manuscript. D.A.G., S.S., requires additional characterization. Notably, while CD4+ T cells are C.R.K., J.R., C.S. and S.D. obtained the funding. D.A.G. and S.D. participated in the conception and design of the study, and wrote the manuscript. primary IL-17 producers, a variety of other cells, including CD8+ T cells and innate lymphoid cells, have been shown to produce IL-17 COMPETING FINANCIAL INTERESTS within the liver58. The authors declare no competing financial interests. Corticosterone levels are higher in mice housed at TS. Our findings Reprints and permissions information is available online at http://www.nature.com/ demonstrated that exogenous administration of corticosterone in TN- reprints/index.html. Publisher’s note: Springer Nature remains neutral with regard to housed mice is sufficient to prevent augmented immune responsive- jurisdictional claims in published maps and institutional affiliations. © 2017 Nature America, Inc., part of Springer Nature. All rights reserved. All rights part Nature. of Springer Inc., America, Nature © 2017 ness. In contrast, adrenalectomy removes the inhibition of NAFLD 1. Ma, C. et al. NAFLD causes selective CD4+ T lymphocyte loss and promotes pathogenesis associated with TS, as compared to TN, housing. These hepatocarcinogenesis. Nature 531, 253–257 (2016). data suggest that specific signaling mediators released via the adrenal 2. Rinella, M.E. Will the increased prevalence of nonalcoholic steatohepatitis (NASH) in the age of better hepatitis C virus therapy make NASH the deadlier disease? glands may play a suppressive role in NAFLD pathogenesis. Of note, Hepatology 54, 1118–1120 (2011). chemical inhibition of -adrenergic receptor signaling exacerbates 3. Tiniakos, D.G., Vos, M.B. & Brunt, E.M. Nonalcoholic fatty liver disease: pathology NAFLD progression in mice59. However, -adrenergic receptors also and pathogenesis. Annu. Rev. Pathol. 5, 145–171 (2010). 4. Rahman, K. et al. Loss of junctional adhesion molecule A promotes severe have a homeostatic role in intestinal, liver and immune cells. Thus, steatohepatitis in mice on a diet high in saturated fat, fructose, and cholesterol. future studies employing the use of 2AR, 3AR and/or GR cell-type- Gastroenterology 151, 733–746 (2016). 5. Mehal, W.Z. The Gordian Knot of dysbiosis, obesity and NAFLD. Nat. Rev. specific knockout are required to better define contribution of these Gastroenterol. Hepatol. 10, 637–644 (2013). pathways to NAFLD pathogenesis. 6. Zhu, L. et al. Characterization of gut microbiomes in nonalcoholic steatohepatitis Human obesity and NAFLD are equally represented in males and (NASH) patients: a connection between endogenous alcohol and NASH. Hepatology 48 57, 601–609 (2013). females . The ability of TN housing to promote the development 7. Miele, L. et al. Increased intestinal permeability and tight junction alterations in of robust obesity and NAFLD in female mice further supports the nonalcoholic fatty liver disease. Hepatology 49, 1877–1887 (2009).

NATURE MEDICINE VOLUME 23 | NUMBER 7 | JULY 2017 837 A RTICLES

8. Vespasiani-Gentilucci, U. et al. Hepatic Toll-like receptor 4 expression is associated with 33. Emre, Y. et al. Mitochondria contribute to LPS-induced MAPK activation via portal inflammation and fibrosis in patients with NAFLD. Liver Int. 35, 569–581 (2015). uncoupling protein UCP2 in macrophages. Biochem. J. 402, 271–278 (2007). 9. Kiziltas, S. et al. TLR4 gene polymorphism in patients with nonalcoholic fatty liver 34. Wahle, M. et al. 2-adrenergic receptors mediate the differential effects of disease in comparison to healthy controls. Metab. Syndr. Relat. Disord. 12, catecholamines on cytokine production of PBMC. J. Interferon Cytokine Res. 25, 165–170 (2014). 384–394 (2005). 10. Mills, K.H. TLR-dependent T cell activation in autoimmunity. Nat. Rev. Immunol. 35. Xu, H. et al. Chronic inflammation in fat plays a crucial role in the development 11, 807–822 (2011). of obesity-related insulin resistance. J. Clin. Invest. 112, 1821–1830 (2003). 11. Harley, I.T. et al. IL-17 signaling accelerates the progression of nonalcoholic fatty 36. Brunt, E.M., Kleiner, D.E., Wilson, L.A., Belt, P. & Neuschwander-Tetri, B.A. liver disease in mice. Hepatology 59, 1830–1839 (2014). Nonalcoholic fatty liver disease (NAFLD) activity score and the histopathologic 12. Rau, M. et al. Progression from nonalcoholic fatty liver to nonalcoholic steatohepatitis diagnosis in NAFLD: distinct clinicopathologic meanings. Hepatology 53, 810–820 is marked by a higher frequency of Th17 cells in the liver and an increased Th17/ (2011). resting regulatory T cell ratio in peripheral blood and in the liver. J. Immunol. 196, 37. Nagaya, T. et al. Down-regulation of SREBP-1c is associated with the development 97–105 (2016). of burned-out NASH. J. Hepatol. 53, 724–731 (2010). 13. Hebbard, L. & George, J. Animal models of nonalcoholic fatty liver disease. Nat. 38. Rizki, G. et al. Mice fed a lipogenic methionine-choline-deficient diet develop Rev. Gastroenterol. Hepatol. 8, 35–44 (2011). hypermetabolism coincident with hepatic suppression of SCD-1. J. Lipid Res. 47, 14. Karp, C.L. Unstressing intemperate models: how cold stress undermines mouse 2280–2290 (2006). modeling. J. Exp. Med. 209, 1069–1074 (2012). 39. Teufel, A. et al. Comparison of gene expression patterns between mouse models of 15. Gordon, C.J. Temperature Regulation in Laboratory Rodents (Cambridge University nonalcoholic fatty liver disease and liver tissues from patients. Gastroenterology Press, 1993). 151, 513–525 (2016). 16. Maloney, S.K., Fuller, A., Mitchell, D., Gordon, C. & Overton, J.M. Translating animal 40. Brown, M.P. et al. Knowledge-based analysis of microarray gene expression data by model research: does it matter that our rodents are cold? Physiology (Bethesda) using support vector machines. Proc. Natl. Acad. Sci. USA 97, 262–267 29, 413–420 (2014). (2000). 17. Stemmer, K. et al. Thermoneutral housing is a critical factor for immune function 41. Alexander, J., Chang, G.Q., Dourmashkin, J.T. & Leibowitz, S.F. Distinct phenotypes and diet-induced obesity in C57BL/6 nude mice. Int. J. Obes. (Lond) 39, 791–797 of obesity-prone AKR/J, DBA2J and C57BL/6J mice compared to control strains. (2015). Int. J. Obes. (Lond) 30, 50–59 (2006). 18. Bowers, S.L., Bilbo, S.D., Dhabhar, F.S. & Nelson, R.J. Stressor-specific alterations 42. Li, L., Chen, L. & Hu, L. Nuclear factor high-mobility group box1 mediating the in corticosterone and immune responses in mice. Brain Behav. Immun. 22, activation of Toll-like receptor 4 signaling in hepatocytes in the early stage of non- 105–113 (2008). alcoholic fatty liver disease in mice. J. Clin. Exp. Hepatol. 1, 123–124 (2011). 19. Giles, D.A. et al. Modulation of ambient temperature promotes inflammation and 43. Wieckowska, A. et al. Increased hepatic and circulating interleukin-6 levels in human initiates atherosclerosis in wild type C57BL/6 mice. Mol. Metab. 5, 1121–1130 nonalcoholic steatohepatitis. Am. J. Gastroenterol. 103, 1372–1379 (2008). (2016). 44. Vandanmagsar, B. et al. The NLRP3 inflammasome instigates obesity-induced 20. Rudaya, A.Y., Steiner, A.A., Robbins, J.R., Dragic, A.S. & Romanovsky, A.A. inflammation and insulin resistance. Nat. Med. 17, 179–188 (2011). Thermoregulatory responses to lipopolysaccharide in the mouse: dependence on the 45. Langrish, C.L. et al. IL-23 drives a pathogenic T cell population that induces dose and ambient temperature. Am. J. Physiol. Regul. Integr. Comp. Physiol. 289, autoimmune inflammation. J. Exp. Med. 201, 233–240 (2005). R1244–R1252 (2005). 46. Ramesh, R. et al. Pro-inflammatory human Th17 cells selectively express P- 21. Moragues, V. & Pinkerton, H. Variation in morbidity and mortality of murine typhus glycoprotein and are refractory to glucocorticoids. J. Exp. Med. 211, 89–104 infection in mice with changes in the environmental temperature. J. Exp. Med. 79, (2014). 41–43 (1944). 47. Sano, T. et al. An IL-23R/IL-22 circuit regulates epithelial serum amyloid A to 22. Foxman, E.F. et al. Temperature-dependent innate defense against the common promote local effector Th17 responses. Cell 163, 381–393 (2015). cold virus limits viral replication at warm temperature in mouse airway cells. Proc. 48. Pan, J.J. & Fallon, M.B. Gender and racial differences in nonalcoholic fatty liver Natl. Acad. Sci. USA 112, 827–832 (2015). disease. World J. Hepatol. 6, 274–283 (2014). 23. Eng, J.W. et al. Housing temperature–induced stress drives therapeutic resistance 49. Kanuri, G. & Bergheim, I. In vitro and in vivo models of non-alcoholic fatty liver in murine tumour models through 2-adrenergic receptor activation. Nat. Commun. disease (NAFLD). Int. J. Mol. Sci. 14, 11963–11980 (2013). 6, 6426 (2015). 50. Bäckhed, F., Manchester, J.K., Semenkovich, C.F. & Gordon, J.I. Mechanisms 24. Tian, X.Y. et al. Thermoneutral housing accelerates metabolic inflammation to underlying the resistance to diet-induced obesity in germ-free mice. Proc. Natl. potentiate atherosclerosis but not insulin resistance. Cell Metab. 23, 165–178 Acad. Sci. USA 104, 979–984 (2007). (2016). 51. Puddu, A., Sanguineti, R., Montecucco, F. & Viviani, G.L. Evidence for the gut 25. Athyros, V.G. et al. Cardiovascular risk across the histological spectrum and the microbiota short-chain fatty acids as key pathophysiological molecules improving clinical manifestations of non-alcoholic fatty liver disease: an update. World J. diabetes. Mediators Inflamm. 2014, 162021 (2014). Gastroenterol. 21, 6820–6834 (2015). 52. Inra, C.N. et al. A perisinusoidal niche for extramedullary haematopoiesis in the 26. Kox, M. et al. Voluntary activation of the sympathetic nervous system and attenuation spleen. Nature 527, 466–471 (2015). of the innate immune response in humans. Proc. Natl. Acad. Sci. USA 111, 53. Alisi, A. et al. Plasma high mobility group box 1 protein reflects fibrosis in pediatric 7379–7384 (2014). nonalcoholic fatty liver disease. Expert Rev. Mol. Diagn. 14, 763–771 (2014). 27. Hanssen, M.J. et al. Short-term cold acclimation improves insulin sensitivity in 54. do Nascimento, J.H., Epifanio, M., Soder, R.B. & Baldisserotto, M. MRI-diagnosed patients with type 2 diabetes mellitus. Nat. Med. 21, 863–865 (2015). nonalcoholic fatty liver disease is correlated to insulin resistance in adolescents. 28. Overton, J.M. Phenotyping small animals as models for the human metabolic Acad. Radiol. 20, 1436–1442 (2013). syndrome: thermoneutrality matters. Int. J. Obes. (Lond) 34 (Suppl. 2), S53–S58 55. Sorrentino, P. et al. Predicting fibrosis worsening in obese patients with NASH (2010). through parenchymal fibronectin, HOMA-IR, and hypertension. Am. J. Gastroenterol. 29. Swoap, S.J. et al. Vagal tone dominates autonomic control of mouse heart rate at 105, 336–344 (2010). thermoneutrality. Am. J. Physiol. Heart Circ. Physiol. 294, H1581–H1588 56. Jin, W. & Dong, C. IL-17 cytokines in immunity and inflammation. Emerg. Microbes (2008). Infect. 2, e60 (2013). 30. Emre, Y. & Nübel, T. Uncoupling protein UCP2: when mitochondrial activity meets 57. Meng, F. et al. Interleukin-17 signaling in inflammatory, Kupffer cells, and hepatic immunity. FEBS Lett. 584, 1437–1442 (2010). stellate cells exacerbates liver fibrosis in mice. Gastroenterology 143, 765–776 31. Bhattacharyya, S., Brown, D.E., Brewer, J.A., Vogt, S.K. & Muglia, L.J. Macrophage (2012). glucocorticoid receptors regulate Toll-like receptor 4–mediated inflammatory 58. Giles, D.A., Moreno-Fernandez, M.E. & Divanovic, S. IL-17 axis driven inflammation

© 2017 Nature America, Inc., part of Springer Nature. All rights reserved. All rights part Nature. of Springer Inc., America, Nature © 2017 responses by selective inhibition of p38 MAP kinase. Blood 109, 4313–4319 in non-alcoholic fatty liver disease progression. Curr. Drug Targets 16, 1315–1323 (2007). (2015). 32. Izeboud, C.A., Mocking, J.A., Monshouwer, M., van Miert, A.S. & Witkamp, R.F. 59. McKee, C. et al. Propranolol, a -adrenoceptor antagonist, worsens liver injury in Participation of -adrenergic receptors on macrophages in modulation of LPS- a model of non-alcoholic steatohepatitis. Biochem. Biophys. Res. Commun. 437, induced cytokine release. J. Recept. Signal Transduct. Res. 19, 191–202 (1999). 597–602 (2013).

838 VOLUME 23 | NUMBER 7 | JULY 2017 NATURE MEDICINE ONLINE METHODS expression was determined through two-way ANOVA with an FDR-corrected Mice. With the exception of AKR mice (Jackson), all male and female mice P-value cutoff of 0.05 and a fold change requirement of >2. For pathway analysis, used were on a C57BL/6 background. WT, Tlr4 fl/fl (ref. 60), Il17ra−/− (Amgen)11 the database at http://toppgene.cchmc.org/ was employed, which amasses onto- and Il17a−/− (ref. 61) mice were bred at Cincinnati Children’s Hospital Medical logical data from over 30 individual repositories63. Center (CCHMC) in a specific-pathogen-free (spf) facility maintained at We performed candidate gene prioritization on differentially regulated genes 22 °C with free access to autoclaved chow (LAB Diet #5010; calories provided by between TN and TS animals on a HFD. Using the Toppgene Suite Candidate carbohydrates (58%), fat (13%) and protein (29%)) and water. Adrenalectomized Prioritization tool, we ranked genes based on their functional similarity mice aged 6 weeks on a C57BL/6 background were obtained from Jackson. For to NASH-related genes (extracted from the NASH-profiler; healthy obese experimental purposes, mice were randomly allocated to groups and housed in versus NASH genes identified by an FDR-corrected P value of < 0.05 (two- separate spf units contained within the same animal barrier but maintained at way ANOVA) and fold change >2). The top 20 ranked genes were submitted either 22 °C or 30–33 °C. All care was provided in accordance with the Guide to ToppCluster and Cytoscape for ontological assessment and visualization for the Care and Use of Laboratory Animals. All studies were approved by the (data depicted in Supplementary Fig. 3g). CCHMC IACUC. Sample size was determined by employing power size cal- Further, using a human NASH cohort, we assessed the ability of genes dif- culations based on the results of preliminary data. Blinding was only used for ferentially regulated under HFD conditions as compared to chow in TS and histological scoring of samples. TN housing conditions to discriminate between healthy obese controls and individuals with NASH. Using support vector machines (SVMs), we quantified Corticosterone levels. Serum corticosterone levels were detected by ELISA the discriminative capabilities of the following gene sets: (1) genes differen- according to the manufacturer’s instructions (Arbor Assays). tially regulated between HFD and chow diets in TN; (2) genes differentially regulated between HFD and chow diets in TS; and (3) differentially regulated qRT–PCR. Tissue samples were homogenized in TRIzol (Invitrogen), genes between healthy obese controls and NASH identified through the NASH- and RNA was then extracted, reverse transcribed to cDNA (Verso cDNA profiler39. Model accuracy was compared between the three gene sets (data Synthesis Kit, Thermo Scientific) and subjected to qPCR analysis (Light depicted in Fig. 2p). Cycler 480 II, Roche); all steps were performed according to the manufacturer’s instructions as previously described62. The following primer pairs were Cytokine production. Cytokines were detected employing biotinylated cap- used: Nr3c1 Forward (For) CCATAATGGCATACCGAAGC Reverse (Rev) ture antibodies, detection antibodies and recombinant protein mouse stand- AGGCCGCTCAGTGTTTTCTA; Adbr2 For TAGCGATCCACTGCAATCAC ards. For in vivo production, in vivo cytokine capture assays were employed Rev ATTTTGGCAACTTCTGGTGC; Ppargc1a For CTGCTAGCAA as previously described11,64,65. Briefly, cytokines were detected using IVCCA GTTTGCCTCA Rev AGTGGTGCAGTGACCAATCA; Ucp1 For employing biotinylated capture antibodies (IL-6 (MP5-32C11), TNF (TN3- TCAGCTGTTCAAAGCACACA Rev GTACCAAGCTGTGCGATGTC; Ucp2 For 19)) detection antibodies and recombinant protein mouse standards, all from TCCTGCTACCTCCCAGAAGA Rev CTGAGACCTCAAAGCAGCCT; Srebp1c eBioscience. Biotinylated capture antibodies were injected via the tail vein or For CTGTCTCACCCCCAGCATAG Rev GATGTGCGAACTGGACACAG; intraperitoneally, and terminal serum collection was performed 24 h later. For Ppara For CATGGGGAGAGAGGACAGA Rev AGTTCGGGAAC in vitro production, ELISAs were employed as previously described11,64. For AAGACGTTG; Col1a1 For GGTTTCCACGTCTCACCATT Rev mice, cytokine levels of TNF and IL-6 were determined according to the manu- ACATGTTCAGCTTTGTGGACC; Col1a2 For AGCAGGTCCTTGGAAACCTT facturer’s protocol (BD OptEIA). Rev AAGGAGTTTCATCTGGCCCT; Ccl2 For AGATGCAGTTAACGCCCCAC Rev TGTCTGGACCCATTCCTTCTTG; Ccl3 For ACCATGACA Cell culture and in vitro cytokine production. Mouse BMDCs were differenti- CTCTGCAACCAAG Rev TTGGAGTCAGCGCAGATCTG; Cxcl1 For ated as previously described66. Mouse splenocytes and BMDCs were stimulated ACCCAAACCGAAGTCATAGC Rev TCTCCGTTACTTGGGGACAC; Il23a with ultrapure LPS (100 ng/ml; Invivogen) for 4 h or with 1M norepinephrine For AGGCTCCCCTTTGAAGATGT Rev TTGTGACCCACAAGGACTCA; or dexamethasone as noted. Saa1 For CATTTGTTCACGAGGCTTTCC Rev GTTTTTCCAGTTAGCTT CCTTCATGT; Saa2 For TGTGTATCCCACAAGGTTTCAGA Rev TTATTA Obesity and the NAFLD model. At 6 weeks of age, mice were randomly sepa- CCCTCTCCTCCTCAAGCA; Actb For GGCCCAGAGCAAGAGAGGTA Rev rated into TS or TN housing facilities. After 2 weeks of acclimation, mice were GGTTGGCCTTAGGTTTCAGG. mRNA expression of each gene was compared fed either an irradiated HFD (Research Diets D12492; 60% of calories from to -actin expression (an endogenous housekeeping gene control). fat) or a chow diet. All food was replaced weekly to avoid contamination. All mice were fasted overnight before glucose metabolism testing, insulin toler- RNA sequencing and gene expression quantification. PBMCs were collected ance testing or terminal harvest (completed from 7–10 a.m.). Glucose and insu- from mice fed a chow diet or a HFD for 8 weeks. Liver samples were collected lin tolerance tests were done as previously described11. Briefly, following an from mice fed a chow diet or a HFD for 24 weeks. Gene expression was deter- overnight fast, glucose tolerance levels were determined by injecting mice with mined by running 50-bp single-end reads (~20 million reads per sample). All 10 l of a 10% dextrose solution per gram of bodyweight, and glucose levels transcriptomic analyses were performed in StrandNGS. Following the removal were quantified kinetically at the times shown. For insulin tolerance testing, of barcodes and primers, raw reads were aligned to the mm10 genome using mice were given 10 l of a 0.15 U/ml solution of insulin (Novolin) per gram o annotations provided by UCSC with the following parameters: (1) minimum f bodyweight. Hepatic triglyceride deposition and serum alanine transami- © 2017 Nature America, Inc., part of Springer Nature. All rights reserved. All rights part Nature. of Springer Inc., America, Nature © 2017 percent identify = 90; (2) maximum percent gaps = 5; (3) minimum aligned nase levels were quantified as previously described11. NAFLD activity score read length = 25; (4) number of matched to output per read = 1; and (5) ignore (NAS) was determined from H&E staining by a certified pathologist accord- reads with more than 5 matches. The proprietary aligner (COBWeb) is based on ing to standard practice36. Total body fat and lean and water mass were deter- the Burrows–Wheeler Transform method. Aligned reads were used to compute mined by nuclear magnetic resonance (Whole Body Composition Analyzer, reads per kilobase per million (RPKM) using the expectation–maximization Echo MRI)67. algorithm for the maximum-likelihood estimation of expression. Further, RPKM values were thresholded at 1 and normalized using the DESeq algorithm, which Histology. Oil red O staining was performed on 5-m flash-frozen tissue computes a normalization factor (NF) for each sample. Within each sample, each sections. Masson’s Trichrome and H&E staining were performed on paraffin- transcript is divided by that transcript’s geometric mean across samples. The embedded tissue blocks. Fibrosis quantification employed the use of an Aperio within-sample median of these values is that sample’s NF. To obtain normalized AT2 Slide Scanner, ImageScope (v. 12.3.1.5011), Color Deconvolution (v9) and counts, a sample’s raw RPKM values are divided by that sample’s NF. Finally, nor- Positive Pixel Count (v9; all Leica). Color deconvolution was used to identify malized per-transcript RPKM values were baselined to the median of all samples. the value of the positive stain. This value (0.622712) was subsequently identified Reasonably expressed transcripts (raw RPKM >3 in 100% of samples in at least in the slides using positive pixel count and percentage positive area, which took one condition) were included for differential expression analysis. Differential into account the total number of pixels identified, was quantified.

doi:10.1038/nm.4346 NATURE MEDICINE Flow cytometry. Single-cell suspensions were derived from hepatic homogenate Microbiome manipulation. For antibiotic-mediated depletion of Gram-negative and stained with directly conjugated monoclonal antibodies or isotype controls. bacteria, a cocktail of neomycin and polymixin B sulfate (0.5 and 0.125 mg/ml, Staining for cytokine expression was completed after 4 h of stimulation with respectively) was added to the drinking water after 8 weeks of HFD feeding. 50 ng/ml phorbol 12-myristate 13-acetate (PMA; Promega), 1 g/ml ionomycin (Calbiochem) and 10 g/ml brefeldin A (Gold Bio). Data collection and analysis Statistical analysis. Sample sizes were determined based on preliminary data, were done as previously described11,62. Briefly, cells were stained with Live/Dead which with respect to obesity and NAFLD modeling included weight gain, stain (Zombie UV Dye, Biolegend) and with directly conjugated monoclonal hepatic triglyceride deposition, immune cell infiltration and hepatocellular antibodies CD45-AF700 (104), CD11b-PE (M1/70), F4/80-APCef780 (BM8), damage. Statistical tests were employed for all data sets with similar variance. Ly6C-Percp (HK1.4), Gr1-FITC (RB6-8C5), NK1.1-PECy7 (PK136), TCR--PE The test used was dependent on both the number of groups being compared (H57-597), CD4-APCef780 (GK1.5), CD8-ef450 (53-6.7), TNF-BV650 (MP6- and whether the data were normally distributed. For normally distributed data, XT22), IL-17A-Percp (17B7) and IL-17F-PE (18F10) (all antibodies from eBio- Student’s t-test was used when the comparison was two groups, while one-way science). Flow cytometry data were then collected using an LSR Fortessa (BD) ANOVA was employed for three or more groups, with Tukey’s post hoc test to flow cytometer and analyzed using FlowJo X software (vX0.7). assess differences between specific groups. For non-parametric data sets, the Mann–Whitney test was employed. Statistical analysis was completed using Bacterial translocation. Liver tissue was homogenized in enriched thioglycol- Prism 5a (GraphPad Software, Inc.). All values are represented as means + late medium and subsequently cultured on TSA II/5% sheep blood plates (BD s.e.m. No animals were excluded from analysis. Studies were not blinded unless Bioscience). otherwise noted above.

Intestinal permeability. 1 cm of freshly isolated jejunum was mounted in a Data availability. Raw data can be accessed at the Gene Expression Omnibus U2500 dual Ussing chamber (Warner Instruments). Transepithelial resistance (GEO) under accession GSE80976. and FITC-dextran flux were determined as previously described68.

Microbiome analysis. Bacterial DNA was isolated from fecal material when 60. McAlees, J.W. et al. Distinct Tlr4-expressing cell compartments control neutrophilic mice arrived at the facility (−2 weeks), before they were placed on a cer- and eosinophilic airway inflammation. Mucosal Immunol. 8, 863–873 (2015). tain diet (0 weeks), and at 12 and 24 weeks after either diet was introduced. 61. Komiyama, Y. et al. IL-17 plays an important role in the development of experimental Partial 16S rRNA gene sequences were amplified, targeting the hypervariable autoimmune encephalomyelitis. J. Immunol. 177, 566–573 (2006). regions v1/v2, using primers 27f (AGAGTTTGATCCTGGCTCAG) and 338r 62. Giles, D.A. et al. Regulation of inflammation by IL-17A and IL-17F modulates non-alcoholic fatty liver disease pathogenesis. PLoS One 11, e0149783 (2016). (TGCTGCCTCCCGTAGGAGT). Equimolar amounts of all samples were sub- 63. Chen, J., Bardes, E.E., Aronow, B.J. & Jegga, A.G. ToppGene Suite for gene list jected to sequencing using a MiSeq sequencer from Illumina. Data were then enrichment analysis and candidate gene prioritization. Nucleic Acids Res. 37, processed using mothur software69 to determine phylotypes and operational W305–W311 (2009). taxonomic units (OTUs) and subjected to statistical analysis. LEfSe analysis was 64. Divanovic, S., Trompette, A., Ashworth, J.I., Rao, M.B. & Karp, C.L. Therapeutic enhancement of protective immunity during experimental leishmaniasis. PLoS Negl. performed using the online tool at https://huttenhower.sph.harvard.edu/gal- Trop. Dis. 5, e1316 (2011). axy/ (ref. 70). For comparison to the human microbiome, individual reads were 65. Finkelman, F., Morris, S., Orekhova, T. & Sehy, D. The in vivo cytokine capture assigned to taxonomies using the QIIME package71. Similarly, raw human 16S assay for measurement of cytokine production in the mouse. Curr. Protoc. Immunol. sequencing data published by Zhu et al.6 were downloaded from http://metagen- 54, 6.28 (2003). 66. Divanovic, S. et al. Negative regulation of Toll-like receptor 4 signaling by the Toll- omics.anl.gov/mgmain.html?mgpage=project&project=mgp1195 and analyzed like receptor homolog RP105. Nat. Immunol. 6, 571–578 (2005). using the same QIIME assignment method and analytic pipeline as were used 67. Castañeda, T.R. et al. Metabolic control by S6 kinases depends on dietary lipids. for the mouse 16S read data. Taxonomic data were first analyzed using Mothur PLoS One 7, e32631 (2012). software to determine significantly different OTUs. Similarly, GraphPad Prism 68. Wu, D. et al. Interleukin-13 (IL-13)/IL-13 receptor alpha1 (IL-13R1) signaling regulates intestinal epithelial cystic fibrosis transmembrane conductance regulator (ver 6.07) was used to plot relative species abundance and compare groups. The channel-dependent Cl− secretion. J. Biol. Chem. 286, 13357–13369 (2011). Mann–Whitney test was applied to the data to determine the statistical signifi- 69. Schloss, P.D. et al. Introducing mothur: open-source, platform-independent, cance of observed differences. Between-class principal-component analysis was community-supported software for describing and comparing microbial communities. performed using the ade4 package (ver 5.12) in R; this software is available upon Appl. Environ. Microbiol. 75, 7537–7541 (2009). 70. Segata, N. et al. Metagenomic biomarker discovery and explanation. Genome Biol. request from the corresponding author. Mouse microbiome raw 16S rRNA data 12, R60 (2011). can be accessed at http://metagenomics.anl.gov/mgmain.html?mgpage=projec 71. Caporaso, J.G. et al. QIIME allows analysis of high-throughput community sequencing t&project=mgp18319. data. Nat. Methods 7, 335–336 (2010). © 2017 Nature America, Inc., part of Springer Nature. All rights reserved. All rights part Nature. of Springer Inc., America, Nature © 2017

NATURE MEDICINE doi:10.1038/nm.4346 ERRATA

Erratum: Thermoneutral housing exacerbates nonalcoholic fatty liver disease in mice and allows for sex-independent disease modeling Daniel A Giles, Maria E Moreno-Fernandez, Traci E Stankiewicz, Simon Graspeuntner, Monica Cappelletti, David Wu, Rajib Mukherjee, Calvin C Chan, Matthew J Lawson, Jared Klarquist, Annika Sünderhauf, Samir Softic, C Ronald Kahn, Kerstin Stemmer, Yoichiro Iwakura, Bruce J Aronow, Rebekah Karns, Kris A Steinbrecher, Christopher L Karp, Rachel Sheridan, Shiva K Shanmukhappa, Damien Reynaud, David B Haslam, Christian Sina, Jan Rupp, Simon P Hogan & Senad Divanovic Nat. Med.; doi:10.1038/nm.4346; corrected online 21 June 2017

In the version of this article initially published online, a grant supporting the authors’ work was omitted from the Acknowledgments section. The grant “NIH T32AI118697 (associated with D.A.G.)” has now been added. The error has been corrected in the print, PDF and HTML versions of this article. © 2017 Nature America, Inc., part of Springer Nature. All rights reserved. All rights part Nature. of Springer Inc., America, Nature © 2017 DOI: 10.1002/JLB.3MA0717-274RR

ARTICLE

Differential outcomes of TLR2 engagement in inflammation-induced preterm birth

Monica Cappelletti1∗ Matthew J. Lawson1,2 Calvin C. Chan1,3,4 Adrienne N. Wilburn1,4 Senad Divanovic1

1Division of Immunobiology, Cincinnati Chil- Abstract dren's Hospital Medical Center, and the Univer- sity of Cincinnati College of Medicine, Cincinnati, Preterm birth (PTB) is the leading cause of neonatal mortality worldwide. Infection and inflam- Ohio, USA (Email: [email protected]) mation are considered main causes of PTB. Among multiple pathogens, Gram-positive bacteria 2 Molecular, Cellular and Biochemical Phar- are commonly linked with induction of PTB. Although activation of innate immune responses, via macology Graduate Program, University of TLR2 engagement, by Gram-positive bacteria is a likely cause, whether induction of PTB depends Cincinnati College of Medicine, Cincinnati, Ohio, USA on the potency of specific microbial components to induce Toll-like receptor (TLR)2-driven inflam- 3Medical Scientist Training Program, University mation has not been elucidated. Here, we show that TLR2 activation by synthetic lipopeptides, of Cincinnati College of Medicine, Cincinnati, Pam2Cys, and Pam3Cys specifically, variably influenced inflammation and subsequent induction Ohio, USA of PTB. Pam2Cys challenge, compared to Pam3Cys, induced PTB and promoted significantly 4Immunology Graduate Program, Cincinnati higher expression of inflammatory cytokines, specifically IL-6 and IFN-!,bothin vivo and in vitro. Children's Hospital Medical Center and the University of Cincinnati College of Medicine, Notably, antibody-mediated neutralization of IL-6 or genetic deletion of type I IFN receptor Cincinnati, Ohio, USA (IFNAR) was sufficient to protect from Pam2Cys-driven PTB and to temper excessive proinflam- ∗Current address: Division of Neonatology and matory cytokine production. Conversely, IFN-! or IL-6 was not sufficient to promote induction of Developmental Biology, David Geffen School of PTB by Pam3Cys. In summary, our data implies a divergent function of TLR2-activating lipopep- Medicine at UCLA, Mattel Children'sHospital tides in the magnitude and type of ligand-driven inflammatory vigor in induction of PTB. UCLA, CA, USA. KEYWORDS Inflammation, TLR2, Preterm Birth, IL-6, IFN-b

1 INTRODUCTION associated with PTB.6 Of note, Listeria species and GBS are pathogens that drive robust induction of proinflammatory cytokines including, IL- 6–11 Preterm birth (PTB) is a leading cause of infant morbidity and mor- 6, TNF, IFN-!,andIFN-". Ureaplasma species (Ureaplasma parvum tality worldwide.1,2 Although the etiology of PTB is multifactorial and and Ureaplasma urealyticum)areGram-positivebacteriacommonlyiso- remains largely enigmatic, it is well accepted that maternal inflamma- lated from the female reproductive tract and in the amniotic fluid of tion, driven by infectious or noninfectious triggers, can play a critical pregnant women.12 However, the contribution of Ureaplasma species role in preterm labor.3 In fact, inflammation and immune activation to PTB is somewhat controversial.12–15 Size variation of the surface- are paramount for uterine activation and the onset of labor.2 Pregnant exposed lipopeptides on Ureaplasma species has been proposed to women are more susceptible to infection by multiple pathogens, which affect its ability to interact with innate immune receptors and impact increases the risk of severe illness and adverse pregnancy outcome.2 consequent potency of immune response.15 Hence, underlying causes Infections by Gram-positive bacteria have the most deleterious out- of the propensity for certain Gram-positive bacterial species to trigger comes during pregnancy.3,4 Listeria monocytogenes,aGram-positive PTB are not well understood. bacteria and common contaminant of a variety of raw foods, has TLRs are a family of evolutionarily conserved innate immune tropism for the placenta and is known to cause PTB in humans and receptors essential for recognition of microbial products.16 Activation 3,5 mice. Moreover, Group B streptococci (GBS), the !-hemolytic, Gram- of TLR signaling by conserved microbial molecular structures leads positive constituent of the normal vaginal microflora, is frequently to cytokine production—via activation of NF-kB, MAPK (JNK, p38,

Abbreviations: GBS, Group B streptococci; EPM, elicited peritoneal macrophage; PAMP, pathogen-associated molecular pattern; PTB, preterm birth;i.p.,intraperitoneal This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. c 2017 The Authors. Society for Leukocyte Biology Published by Wiley Periodicals, Inc. ⃝

Received: 3 July 2017 Revised: 17 October 2017 Accepted: 19 October 2017 JLeukocBiol.2018;103:535–543. www.jleukbio.org 535 536 CAPPELLETTI ET AL.

ERK), and IRF signaling pathways, leading to proinflammatory (e.g., cated concentrations and treated with IL-6-neutralizing antibody TNF-#,IL-8,IL-6,IFN-",andIFN-!), anti-inflammatory (e.g., IL-10), (500 $g/mouse; clone MP5-20F3; BioXCell, West Lebanon, NH) or and immunoregulatory (e.g., IL-12) cytokine production.17,18 TLRs are recombinant mouse IL-6 (eBioscience, San Diego, CA) as indicated. 4 robustly expressed at the maternal/fetal interface by the pregnant Exogenous administration of 10 U/mouse IFN-! (PBL Biomedical uterus, placenta, amniotic membranes, and trophoblast.19–23 In mice, Laboratories, Piscataway, NJ) was performed 4 h prior to Pam3Cys challenge with diverse TLR ligands—via intraperitoneal, intrauterine, and Pam2Cys challenge and IL-6-neutralizing antibody administration. or intraamniotic routes—increases proinflammatory cytokine release PTB was defined as parturition within 24 h after TLR ligand challenge in the uterus5 and fetal membranes,24,25 recruits immune cells into (all pups deceased). Term birth was defined as parturition between the cervix,26,27 and induces PTB.27–31 Notably, TLR2 is involved in the days 19 and 21 of gestation (all pups alive). recognition of lipoteichoic acid and lipopeptides from Gram-positive bacteria. Specifically, the ability of TLR2 to heterodimerize with 2.4 Cytokine quantification either TLR6 or TLR1 results in the recognition of diacylated and In vivo systemic IL-6, TNF, and IFN-" levels were quantified using in vivo triacylated lipopeptides, respectively, hinting at a potential mechanism cytokine capture assay.5,32,33 Briefly, biotinylated capture antibodies to discriminate among various microbial ligands and to elicit varied (eBioscience) were injected i.p. 3 h prior to Pam3Cys and Pam2Cys downstream inflammatory responses. However, the contribution of administration and serum cytokine levels were determined 4 h later. specific TLR2-activating lipopeptide species in the context of PTB has In vitro murine thioglycollate elicited peritoneal macrophages not been examined. (EPMs) were generated using standard protocol.5,34 EPMs (1 × 106 In this study, we hypothesized that induction of PTB depends on the cells/well) were cultured with or without IFN-! (250 U/ml) for 4 h, sub- potency of specific lipopeptides to induce TLR2-driven inflammation. sequently mock-stimulated or stimulated with Pam3Cys (1.5 $g/ml) or Our data demonstrate that recognition of diacylated lipopeptide (e.g., Pam2Cys (1.5 $g/ml) for 4 h to determine mRNA expression, or with Pam2Cys) induced PTB and greater proinflammatory cytokine produc- Pam3Cys (100 ng/ml) or Pam2Cys (100 ng/ml) for 18 h to determine tion compared to triacylated lipopeptide (e.g., Pam3Cys). Our data also cytokine production by ELISA (BD Biosciences, San Diego, CA). suggest that sufficiency of inflammatory mediators to induce PTB following Pam2Cys or Pam3Cys challenge is varied. In sum, these data 2.5 Type I IFN activity quantification argue that TLR2-driven induction of PTB might depend on the magnitude and milieu of ligand-driven inflammatory vigor and such dif- TypeIIFNactivityinmouseserumsamplesorcellculturesupernatants ferences might shed light on why only certain Gram-positive bacterial was measured with reference to a recombinant mouse IFN-! using an species are associated with induction of PTB. L-929 cell line transfected with an IFN-sensitive luciferase construct as previously described.5,35 Luciferase activity was measured on a luminometer (SpectraMax L; Molecular Devices, Sunnyvale, CA). 2 MATERIALS AND METHODS 2.6 Gene expression 2.1 Reagents For quantification of mRNA expression in murine samples, cells/tissues All cell culture reagents were endotoxin free to the limits of detec- were homogenized in TRIzol (Invitrogen, San Diego, CA), RNA was tion of the Limulus amebocyte lysate assay (Lonza, Visp, Switzerland) at extracted, and cDNA was generated and quantified as previously the concentrations employed. All TLR ligands (Pam3Cys and Pam2Cys) described32 using Light Cycler 480 II (Roche, San Francisco, CA). The used were of ultrapure grade (Invivogen, San Diego, CA). following murine primers were utilized: IL-6: TGGTACTCCAGAAGACCAGAGG and AACGATGATGCACTT- 2.2 Mice GCAGA, Animals were housed in a specific pathogen-free animal facility at TNF: CCAGACCCTCACACTCAGATCA and CACTTGGTGGTTTGC- CCHMC and handled in high-efficiency particulate-filtered laminar TACGAC. flow hoods with free access to food and water. All studies were IFNB1: TCCAGCTCCAAGAAAGGACG and TTGAAGTCCGCCCTG- performed in accordance with the procedures outlined in the Guide TAGGT, for the Care and Use of Laboratory Animals and approved by the MX-1: CTCAGGGTGTCGATGAGGTC and TCTGAGGAGAGCCAGAC- CCHMC Institutional Animal Care and Use Committee. Female mice GAT, (WT, IFNAR− /− ), on a C57BL/6J background, were mated with fertile IRF9: CACTCGGCCACCATAG and AAGCCATCTCTTTCCAAGTCTTT, male mice of the same strain. The presence of a vaginal plug was con- ISG15: GTCACGGACACCAGGAAATC and AAGCAGCCAGCCGCA- sidered at day 1 of pregnancy. Parturition events were monitored on GACTG. days 17–21 of gestation and defined as complete delivery of pups. 2.7 Statistical analysis 2.3 Preterm birth Data were analyzed by unpaired Student t test in Prism 5a (GraphPad On day 16 of gestation, gravid female mice were mock-challenged Software Inc., La Jolla, CA) as appropriate. A P value less than 0.05 was (saline) i.p. or challenged with Pam3Cys and Pam2Cys at indi- considered significant. In vivo serum cytokine values were normalized CAPPELLETTI ET AL. 537

AB100 100 D1 D16 D17 D19/21 80 80 Vaginal Pam3Cys PTB Term plug Pam2Cys birth 60 60

40 40

20 20 3/6 6/6 Preterm birth (%) birth Preterm Preterm birth (%) 0/3 0/3 0/3 0 0

100 250 100 200 1000 Pam3Cys (µg/mouse) Pam2Cys (µg/mouse)

C 500 40 15 *** 100 *** 40 *** D 400 80 30 30 10 300 60 (U/ml) β 20 (ng/ml) 20 200 γ 40 IFN- IL-6 (ng/ml) TNF (ng/ml) TNF 5 IFN- 10 (%) birth Preterm 20 100 10 0/3 0/3 0/5 0 0 0 0 0 rIL-6 + +– Saline Saline Saline Saline Pam3Cys –+– Pam3CysPam2Cys Pam3CysPam2Cys Pam3CysPam2Cys Pam3CysPam2Cys Pam2Cys ––+ α-IL-6 ––+

FIGURE 1 Activation of TLR2 by Pam2Cys but not Pam3Cys induces preterm birth. Gravid WT mice challenged with (A) Pam3Cys or (B) Pam2Cys at the indicated doses on day 16 of gestation and the incidence of PTB was quantified. (C) WT mice (n = 4–7/condition) were challenged with saline, Pam3Cys (100 $g/mouse) or Pam2Cys (100 $g/mouse) for 4 h, and serum IL-6, TNF, IFN-",andIFN-! levels were quantified by in vivo cytokine capture assay (IVCCA) and type I IFN activity assay, respectively. (D) Gravid WT mice were treated with IL-6 neutralizing Ab (500 $g/mouse) or recombinant mouse IL-6 (10 $g/mouse) followed by above-described challenge (Figure 1B) and the incidence of PTB was quantified. (A, B and D) Data represent percent of induction of PTB. Student's t test, ***P < 0.001 to Pam2Cys challenge as appropriate and expressed as the percent (%) mice and humans—indicating that mice are a useful experimen- change. All values are represented as means ± SE or as percent of term tal tool for the interrogation of the role of infection/inflammation or PTB induction. in induction of PTB.21,41 Further, in mice following i.p. TLR ligand injection, parturition-associated molecules are upregulated in the entire gestational environment and affect all fetuses.5 Hence, the par- 3 RESULTS AND DISCUSSION turition process once initiated results in the delivery of all pups. Toovercome the complexity of a live bacterial infection (e.g., replica- 3.1 Activation of TLR2 by Pam2Cys but not tion rate, tropism for various cell subsets, activation of multiple innate Pam3Cys induces PTB immune receptors), we used purified synthetic lipopeptides known Gram-positive bacteria (e.g., Listeria, GBS, and Ureaplasma species) to activate TLR2 signaling—pathogen-associated molecular patterns are pathogens commonly associated with PTB.5,6,12–15 However, com- (PAMPs) that molecularly mimic the acylated amino terminus of bac- petency of certain bacteria to trigger PTB remains unclear. Although terial lipopeptides.42 Pregnant mice at day 16 of gestation were pro- all experimental models of PTB have limitations compared to human tected from PTB even at very high doses of triacylated lipopeptide pregnancy and parturition, mice represent the most common ani- (Pam3Cys; 1 mg/mouse) challenge (Figure 1A). Conversely, challenge mal model of PTB.2 Specifically, in such experimental models, the with diacylated lipopeptide (Pam2Cys; 200 $g/mouse) induced 100% doses of inflammatory ligands employed are high. The intent being PTB at the commonly used doses43–45 (Figure 1B). In this experimen- to enable 100% parturition incidence in fully backcrossed mice.5,36–40 tal setting utilizing Pam2Cys challenge, all delivered pups died (either Notably, that allows investigators to uncover differences between in utero or upon delivery) while the mothers survived the challenge animal genotypes and/or various ligands in comparison to estab- without presenting apparent adverse outcomes that could hinder their lished baseline reads in wild-type (WT) mice. Here, we utilized a health status. In mice, lung maturation is not fully completed at day 16 well-established, tractable, model of PTB induction in mice5 for of gestation.46 As our model invokes initiation of inflammatory insult mechanistic insight into how early parturition is triggered in the at day 16 of gestation, the lack of lung maturation/function in prema- context of systemic TLR2 challenge (Figure 1A and B). The path- turely born pups may represent the likely locus responsible for fetal ways regulating immune responses are highly conserved between death upon premature parturition.36 538 CAPPELLETTI ET AL.

As inflammation is directly linked with TLR-driven induction of findings invoking divergent effects of Pam2Cys and Pam3Cys in induc- PTB, the capacity of Pam2Cys and Pam3Cys to induce inflammatory tion of inflammatory vigor and PTB, highlight the relevance of future responses was examined next. Induction of PTB by Pam2Cys corre- studies focused on the role of TLR1 and TLR6 signaling in such bio- lated with significantly increased IL-6, IFN-",andIFN-!,butsimilar logical outcomes. Overall, such factors might posit the specificity of TNF production compared to Pam3Cys challenge—findings when per- certain Gram-positive bacterial species to augment immune responses formed with equivalent doses of these TLR2 ligands (Figure 1C). The and trigger PTB. central role of proinflammatory cytokines, and IL-6 in particular, is supported by the correlation between IL-6 levels and PTB in humans47 3.2 Pam2Cys drives expression of type I IFN and in mouse models of TLR-driven inflammation.5,38 Pharmacologi- signature genes cal inhibition of IL-6 (antibody mediated) was sufficient to protect from Pam2Cys-drivenPTB,whileasingledoseofrIL-6aloneorrIL-6incom- Type I IFNs, rapidly induced in various cell types upon viral and bac- bination with Pam3Cys was not sufficient to trigger induction of PTB terial infections, represent key signaling cytokines central to activa- (Figure 1D)—something supported by a previously published report.48 tion of both innate and adaptive immune responses. All type I IFNs Such findings suggest that Pam2Cys and Pam3Cys signaling activate bind a common receptor at the surface of cells, which is known as divergent immune mediators that may play a role in induction of PTB. the type I IFN receptor. The type I IFN receptor is composed of 2 Further, these data also suggest that IL-6 is not the only proinflamma- subunits, IFNAR1 and IFNAR2, which are associated with TYK2 and tory cytokine upregulated in the context of PTB and that other immune JAK1, respectively.59 The outcomes of Gram-positive bacterial lig- mediators likely play a role in induction of PTB. Specifically, of interest and dependent activation of TLR2 signaling are commonly focused on to our findings, published reports suggest a role for IFN-" in the regu- proinflammatory cytokine production. However, recent studies have lation of mediators involved in the onset of labor.49 Importantly, such demonstrated that, in addition to NF-kB-dependent proinflammatory findings warrant that future studies should expand the knowledge of cytokines,60 intracellular TLR2 activation by viral and bacterial com- IFN-" in inflammation-driven PTB. ponents (e.g., UV irradiated and live virus, Pam2Cys, Pam3Cys, and Overall, our data suggest that Pam2Cys mediates a more potent MALP-2, respectively) also leads to the production of IFN-!-and TLR2-driven inflammatory response than Pam3Cys and that IL-6 type I IFN-dependent responses.45,60,61 Of note, the type I IFN axis induction by Pam2Cys, but not Pam3Cys, plays an important role plays a critical role in TLR-driven PTB and exacerbation of systemic in PTB. Enhanced solubility of Pam2Cys in saline—something that and reproductive site inflammation.5 Hence, we examined the robust- is dependent not only on the lipid content but also on the posi- ness of TLR2-activating lipopeptides to promote IFN-! production and tion of attachment of the lipids50 and a more potent stimulation of expression of type I IFN signature genes. 51,52 splenocytes and macrophages compared to Pam3Cys might be In vivo stimulation by Pam2Cys led to increased IFN-! production central to the observed effects. Environmental conditions including compared to Pam3Cys (Figure 1C). Similarly, in vitro stimulation of peri- acidic pH, a postlogarithmic bacterial growth phase, high tempera- toneal macrophages by Pam2Cys triggered a more robust expression tures, and high salt concentrations are all known to support the accu- of IFN!1andtypeIIFNsignaturegenes(e.g.,Mx1,Isg15,andIRF9) mulation of the diacyl lipopeptides. Hence, such conditions may pos- (Figure 2A). Further, this increase was concomitant with enhanced IL- sibly impact the balance between di- versus triacylated lipopeptides 6andsimilarTNFmRNAlevelscomparedtoPam3Cys(Figure2B). expression by certain bacteria and influence the type and vigor of Although the type I IFN family in mice is composed of IFN-# (with 53 immune responses. Whether these conditions are present in the 14 known subtypes) and IFN-!,TLR2-driventypeIIFN-dependent intrauterine environment and specifically in the placenta and amni- responses are believed to be largely dependent on the activity of IFN- 60 otic fluid during pathological pregnancies, however, needs to be fur- !. Notably, in agreement with the published report, IFN-!-deficient ther elucidated. It is plausible that the pregnant uterus might repre- macrophages exhibit reduced IL-6 expression following Pam2Cys and sent a favorable environment for the replication of certain pathogens Pam3Cys stimulation in vitro (data not shown). Moreover, we and oth- and might regulate the balance of lipopeptides expression of PTB- ers have shown that dysregulation of IFN-! is a major determinant for related bacteria, thus directly impacting the type of inflammatory PTB in context of a polymicrobial infection.5,62 However, the possible 54 responses by both immune and nonimmune cells. Of note, L. monocy- contribution of other type I IFNs (e.g., IFN-# isotypes) has not been togenes and Mycoplasma fermentans,pathogenscorrelatedwithPTBin formally examined. humans, exclusively contains/expresses diacylated lipoproteins, struc- Although TLR2 activation was thought to specifically induce NF- turally similar to Pam2Cys, in its cell wall.55,56 Similarly, U. parvum, kB-driven cytokine production, recently the contribution of endolyso- associated with PTB in humans, similarly express diacylated lipopep- somal TLR2 activation has been shown to promote type I IFN tide and are essential for the initiation of inflammation.57 Further, the production.60 However, the contribution of TLR2 homodimerization or differential ability of lipopeptides to induce inflammation might rely dimerization of TLR2 with either TLR1 or TLR6 for induction of type on specific amino acids that affect their potency in inducing TLR2- I IFN is not known.63 Hence, an improved mechanistic understanding dependent responses and on the ability of TLR2 to heterodimerize of TLR2/1 and TLR2/6 heterodimeric signaling and downstream path- with either TLR6 or TLR1.58 However, the contribution of either TLR6 ways in induction of PTB would be of high interest and importance or TLR1 heterodimerization to TLR2 in inflammation-driven PTB has to the research community. Lastly, the activating as well as regulatory not been examined. Hence, these gaps in knowledge, coupled with our mechanisms downstream of TLR2-dependent induction of type I IFNs CAPPELLETTI ET AL. 539

A 3 8 60 4 * 10 6 4 3 2 * 3 4 2 1 2 2 1 MX1 (fold change) change) (fold MX1 IRF9 (fold change) change) (fold IRF9

IFNb1 (fold change) change) (fold IFNb1 1 ISG15 (fold change) change) (fold ISG15

0 0 0 0 β β β β NS NS NS NS IFN- IFN- IFN- IFN- am2Cys Pam3CysPam2Cys Pam3CysPam2Cys Pam3CysP Pam3CysPam2Cys

B 20 * 20

15 15

10 10

5 5 IL-6 (fold change) change) IL-6 (fold TNF (fold change) change) (fold TNF

0 0 β β NS NS IFN- IFN- Pam3CysPam2Cys Pam3CysPam2Cys

FIGURE 2 Pam2Cys challenge promotes type I IFN-associated gene expression. WT murine peritoneal macrophages were treated with Pam2Cys (1.5 $g/ml), Pam3Cys (1.5 $g/ml), and IFN-! (250 U/ml) for 4 h and expression levels of (A) type I IFN signature genes (IFN-!1, Mx1, Isg15, and Irf9) and (B) proinflammatory cytokines (IL-6 and TNF) were quantified by PCR. Data are representative of 3 independent experiments. Data are normalized to not stimulated (NS) condition need to be elucidated. Thus, TLR2 signaling potentially discriminates activation, macrophage infiltration and activation, and uterine tissue avarietyofpathogensandregulatesthenatureofresponsesagainst inflammation (e.g. Cox-2 and prostaglandin E2)maysimilarlyplaya 5,66–69 both extracellular and intracellular microbes. role. The contribution of IFN-# subtypes and IFN-!,something that is pathogen specific as well as anatomical locus/tissue/cell type 59 3.3 IFNAR signaling contributes to specific, also remains under defined. However, IFN-! might have Pam2Cys-mediated induction of PTB apredominantroleasitisanimportantdeterminantofpathogen- related virulence and is robustly induced by Gram-positive bacteria Given the robust induction of type I IFN production and signature associated with PTB (e.g. Listeria species, GBS).7,60 after Pam2Cys-driven TLR2 activation, the necessity of the common IFN / receptor (IFNAR) in proinflammatory cytokine production and # ! 3.4 IFN-!-driven enhancement of PTB was further investigated.64 IFNAR is a heteromeric cell surface Pam3Cys-associated inflammation is not sufficient for receptor with 2 subunits, IFNAR1 and IFNAR2. Of note, IFN- specif- ! induction of PTB ically interacts with IFNAR1 in an IFNAR2-independent manner.65 Compared to WT-treated mice, genetic deletion of IFNAR1 tempered Acombinedrecognitionofviralandbacterialmolecularpatterns Pam2Cys-dependent induction of IL-6 and TNF (Figure 3A and B) by TLR, coined a “double hit hypothesis”, has been proposed as and resulted in a 50% protection from Pam2Cys-driven PTB (Figure acentralregulatorofincreasedsusceptibilitytoPTB.40,70,71 We 3C). The lack of IFNAR1 signaling similarly decreased Pam3Cys-driven have previously demonstrated that type I IFNs regulates TLR4- IL-6 and TNF production. Further Pam2Cys-driven cytokine produc- driven inflammatory vigor, and specifically primes for exacerbated tion was significantly higher compared to Pam3Cys in both WT and IL-6 and TNF production.5 As Gram-positive bacteria-derived PAMPs IFNAR− /− mice (Figure 3A and B). These data suggest that both syn- are associated with PTB and robust activation of the type I IFN thetic lipopeptides activate the type I IFN/IFNAR axis and that such axis, we asked whether type I IFN priming acts as a universal sen- activation regulates proinflammatory cytokine production in this con- sitization mechanism to secondary challenge with all TLRs (includ- text. However, although significant, the protection from Pam2Cys- ing those that alone do not induce PTB). Of note, in vitro IFN-! driven PTB in IFNAR-deficient mice was not complete (Figure 3C), indi- administration was sufficient, although at different levels, to prime cating that additional signaling pathways activated following Pam2Cys for a secondary challenge with Pam2Cys and Pam3Cys (Figure 4A). challenge are similarly involved. Hence, how IFNAR signaling (e.g., This was recapitulated in vivo as IFN-! priming exacerbated IL-6, IFNAR1 and IFNAR2) contributes to protection from Pam2Cys-driven TNF, and IFN-" production by both Pam2Cys and Pam3Cys. How- PTB warrants further investigation. ever, IFN-! priming induced significantly higher levels of proin- In fact, IFNAR modulation of chemokine production (e.g., CCL2 and flammatory cytokines upon secondary challenge with Pam2Cys CCL4), proinflammatory cytokines (IL-1! and IL-18), inflammasome (Figure 4B and C). 540 CAPPELLETTI ET AL.

-/- WT IFNAR-/- WT IFNAR A 150 150 *** ***

100 100 change) change)

50 ** 50 IL-6 (% IL-6 (% *** change) TNF (%

0 0

Pam2Cys Pam3Cys Pam2Cys Pam3Cys Pam2Cys Pam3Cys Pam2Cys Pam3Cys

C Pam2Cys Pam3Cys Pam2Cys B 150 160 150 150 100

80 120 100 100 100 60

80 40

50 50 50 *** ** 20 TNF (% change) change) (% TNF 3/6 IL-6 (% change) change) (% IL-6 6/6 40 change) (% TNF

IL-6 (% change) change) (% IL-6 *** Pretermbirth (%) * *** 0 0 0 0 0 /- -/- -/- -/- - -/- WT WT WT WT WT

IFNAR IFNAR IFNAR IFNAR IFNAR

FIGURE 3 IFNAR signaling is required for Pam2Cys-mediated proinflammatory cytokine production and induction of preterm birth. (A and / B) WT and IFNAR− − mice (n = 3–7/condition) were challenged with Pam3Cys (100 $g/mouse) or Pam2Cys (100 $g/mouse) for 4 h and serum IL-6 and TNF levels were quantified by in vivo cytokine capture assay (IVCCA). (C) Gravid WT and IFNAR− /− mice were challenged with Pam2Cys (200 $g/mouse) on day 16 of gestation and the incidence of PTB was quantified. In vivo serum cytokine values were normalized to Pam2Cys challenge as appropriate and expressed as the percent (%) change. Student's t test, **P < 0.01, ***P < 0.001

The relevance and sufficiency of such effects on PTB was exam- this report, is limited to certain secondary challenge.5 Further, our data ined next. IFN-! sensitization was not sufficient to augment Pam3Cys- provide novel insights into the mechanisms regulating the ability of dif- driven inflammatory vigor necessary for induction of PTB, even at ferent TLR2 ligands to augment inflammatory vigor and increase the very high concentration of ligand (Figure 4D). These data suggest that risk of PTB. Pam2Cys and Pam3Cys have distinct functional consequences upon Such findings are in agreement with the published reports, sug- the activation of different heterodimers. Whether IFN-! sensitization gesting that diacylated lipopeptides are commonly present in Gram- is sufficient to lower the necessary threshold for Pam2Cys for induc- positive bacterial species correlated with PTB.55–57 Here, we demon- tion of PTB, as observed for LPS,5 still needs to be defined. Differ- strate for the first time that: (1) activation of TLR2 by the synthetic ential contribution of the MyD88- and TRIF-dependent pathways in lipopeptide Pam2Cys, but not Pam3Cys, induced PTB; (2) Pam2Cys TLR2-mediated signal transduction and subsequent induction of gene challenge, compared to Pam3Cys, enhanced expression of IFN-! expression may be responsible for optimal host responses toward cer- and type I IFN signature genes; (3) IFNAR signaling is required for tain infections.72 Furthermore,multiplefactorsareknowntoaffectthe Pam2Cys- and Pam3Cys-mediated proinflammatory cytokine produc- role of type I IFNs in controlling susceptibility to secondary bacterial tion; (4) IFNAR signaling contributes to Pam2Cys-driven induction 5 challenge including limited window of type I IFN priming, anatomical of PTB; and (5) ability of IFN-! priming/sensing is limited to certain locus of sensitization and challenge,59 critical cell types for type I IFN molecular patterns as it was not sufficient to augment Pam3Cys-driven sensitization, and variable mechanisms that limit the duration and the inflammatory vigor and induction of PTB. magnitude of the inflammatory response.73 Our findings invoke several salient questions that remain to be elucidated including: Are specific Gram-positive bacteria more likely to induce PTB because of expression of certain lipopeptides? Can cer- 4 CONCLUSIONS tain lipopeptides induce protective effects by activating mechanisms that regulate overt inflammatory responses? Does TLR2 homodimer- Overall, these data suggest that sufficiency of type I IFN priming in ization or heterodimerization of TLR2 with TLR1 and TLR6 regulate augmentation of inflammatory vigor, at least in conditions utilized in the potency of immune response and induction of PTB? What are CAPPELLETTI ET AL. 541

In vitro In vivo A 120 B 400 ** 400 200 *** *** ** 90 300 300 150 nge) nge)

60 200 200 100 (% change) change) (% γ

TNF (% change) change) TNF (% *** cha IL-6 (% IFN- IL-6 (fold change) change) (fold IL-6 30 100 100 50

0 0 0 0 IFN-β – – + – + IFN-β – + – + – + Pam2Cys – + + – – Pam2Cys + + + + + + Pam3Cys – – – + +

In vivo C 400 400 120 D 100 *** ** 80 300 300 90 60

200 200 60 40 (% change) change) (% γ

** Preterm birth (%) IL-6 (% change) change) IL-6 (% TNF (% change) change) TNF (% 20

100 100 IFN- 30

3/6

0/3 0/4 0/3 0 0 0 0 IFN-β – + – + – + IFN-β + + + Pam3Cys + + + + + + Pam3Cys 100 200 500 (µg/mouse)

FIGURE 4 IFN-!-driven enhancement of Pam3Cys-associated inflammation is not sufficient for induction of preterm birth. (A) WT murine peritoneal macrophages (n = 3) were treated with recombinant mouse IFN-! (250 U/ml; 4 h) prior to being challenged with Pam2Cys (100 ng/ml) and Pam3Cys (100 ng/ml) for 18 h and IL-6 levels were quantified by ELISA. (B and C) WT mice (n = 3–6/condition) were challenged with recombi- 4 nant mouse IFN-! (10 U/mouse for 4 h) prior to being challenged with Pam3Cys or Pam2Cys (25 $g/mouse for 4 h) and serum IL-6 and TNF levels 4 were quantified by in vivo cytokine capture assay (IVCCA). (D) Gravid WT mice were challenged with recombinant mouse IFN-! (10 U/mouse for 4h)priortobeingchallengedwithPam3Cysattheindicateddosesonday16ofgestationandtheincidenceofPTBwasquantified.Cytokinevalues were normalized to (A) not stimulated control and (B) Pam2Cys or (C) Pam3Cys challenge as appropriate and expressed as the percent (%) change. Student's t test, **P < 0.01, ***P < 0.001 the critical TLR2 downstream pathway(s) that regulates the magni- ACKNOWLEDGMENTS tude of inflammation? Do subclinical viral infections predispose to This study was supported, in part, by the Cincinnati Children'sHos- asecondaryGram-positivechallengeandthusincreasetheriskof pital Medical Center (CCHMC) Perinatal Institute Pilot and Feasibil- augmented systemic inflammation and PTB? What is the contribu- ity Award, Burroughs Wellcome Fund Preterm Birth Research Grant tion of fetal immune response to Gram-positive bacteria? In summary, (#1015032), March of Dimes Prematurity Research Center Ohio Col- our novel data shed light on the complex nature of innate immune laborative Innovation Catalyst Grant (associated with S.D.), and NIH activation strength and argue that ability of pathogens to induce PTB, T32AI118697 (associated with C.C.C.). We thank Traci Stankiewicz in an experimental setting, relies on the magnitude and type of TLR- and Maria Moreno-Fernandez for technical assistance. driven inflammation. DISCLOSURES

The authors declare no conflict of interest. AUTHORSHIP

M.C., M.J.L., C.C.C., A.N.W., and S.D. participated in generation, analysis, REFERENCES and interpretation of data. M.C. and S.D. participated in the concep- 1. Romero R, Dey SK, Fisher SJ. Preterm labor: one syndrome, many tion and design of the study and wrote the manuscript. S.D. obtained causes. Science.2014;345:760–765. the funding. All authors have reviewed the manuscript and approve the 2. Cappelletti M, Della Bella S, Ferrazzi E, Mavilio D, Divanovic S. Inflam- final version. mation and preterm birth. JLeukocBiol.2016;99:67–78. 542 CAPPELLETTI ET AL.

3. Kourtis AP, Read JS, Jamieson DJ. Pregnancy and infection. New Engl J 23. Patni S, Flynn P, Wynen LP, et al. An introduction to Toll-like Med.2014;370:2211–2218. receptors and their possible role in the initiation of labour. BJOG. 4. Payne MS, Bayatibojakhi S. Exploring preterm birth as a polymicro- 2007;114:1326–1334. bial disease: an overview of the uterine microbiome. Front Immunol. 24. Schaefer TM, Desouza K, Fahey JV, Beagley KW, Wira CR. Toll-like 2014;5:595. receptor (TLR) expression and TLR-mediated cytokine/chemokine 5. Cappelletti M, Presicce P, Lawson MJ, et al. Type I interferons regu- production by human uterine epithelial cells. Immunology. 2004;112:428–436. late susceptibility to inflammation-induced preterm birth. JCI Insight. 2017;2:e91288. 25. Schaefer TM, Fahey JV, Wright JA, Wira CR. Innate immunity in 6. Surve MV, Anil A, Kamath KG, et al. Membrane vesicles of group B the human female reproductive tract: antiviral response of uterine Streptococcus disrupt feto-maternal barrier leading to preterm birth. epithelial cells to the TLR3 agonist poly(I:C). J Immunol.2005;174: 992–1002. PLoS Pathog.2016;12:e1005816. 7. Stockinger S, Kastner R, Kernbauer E, et al. Characterization of the 26. Yellon SM, Ebner CA, Sugimoto Y. Parturition and recruitment of macrophages in cervix of mice lacking the prostaglandin F receptor. interferon-producing cell in mice infected with Listeria monocytogenes. .2008;78:438–444. PLoS Pathog.2009;5:e1000355. Biol Reprod 8. HogeJ,YanI,JannerN,etal.IL-6controlstheinnateimmuneresponse 27. Norman JE. Preterm labour. Cervical function and prematurity. Best .2007;21:791–806. against Listeria monocytogenes via classical IL-6 signaling. J Immunol. Pract Res Clin Obstet Gynaecol 2013;190:703–711. 28. Waldorf Adam, M K, Persing D, Novy MJ, Sadowsky DW, Gravett MG. Pretreatment with Toll-like receptor 4 antagonist inhibits 9. Cao L, Lawrence DA. Suppression of host resistance to Listeria monocy- lipopolysaccharide-induced preterm uterine contractility, cytokines, togenes by acute cold/restraint stress: lack of direct IL-6 involvement. J and prostaglandins in rhesus monkeys. .2008;15: Neuroimmunol.2002;133:132–143. Reprod Sci 121–127. 10. Torres D, Janot L, Quesniaux VF, et al. Membrane tumor necrosis 29. Elovitz MA, Wang Z, Chien EK, Rychlik DF, Phillippe M. A new factor confers partial protection to Listeria infection. Am J Pathol. 2005;167:1677–1687. model for inflammation-induced preterm birth: the role of platelet- activating factor and Toll-like receptor-4. Am J Pathol.2003;163: 11. Clarke D, Letendre C, Lecours MP, et al. Group B Streptococcus 2103–2111. induces a robust IFN-gamma response by CD4(+)Tcellsinaninvitro 30. Ilievski V, Lu SJ, Hirsch E. Activation of Toll-like receptors 2 or 3 and and in vivo model. J Immunol Res.2016;167:5290604. preterm delivery in the mouse. Reprod Sci.2007;14:315–320. 12. Novy MJ, Duffy L, Axthelm MK, et al. Ureaplasma parvum or Mycoplasma hominis as sole pathogens cause chorioamnionitis, 31. Koga K, Cardenas I, Aldo P, et al. Activation of TLR3 in the trophoblast is associated with preterm delivery. . preterm delivery, and fetal pneumonia in rhesus macaques. Reprod Sci. Am J Reprod Immunol 2009;16:56–70. 2009;61:196–212. 13. Combs CA, Gravett M, Garite TJ, et al., Amniotic fluid infection, inflam- 32. Harley IT, Stankiewicz TE, Giles DA, et al. IL-17 signaling accelerates the progression of nonalcoholic fatty liver disease in mice. . mation, and colonization in preterm labor with intact membranes. Am Hepatology 2014;59:1830–1839. JObstetGynecol.2014;210:125e1–125e15. 14. Senthamaraikannan P, Presicce P, Rueda CM, et al. Intra-amniotic 33. Giles DA, Moreno-Fernandez ME, Stankiewicz TE, et al. Thermoneu- ureaplasma parvum-induced maternal and fetal inflammation and tral housing exacerbates nonalcoholic fatty liver disease in mice and allows for sex-independent disease modeling. .2017;23: immune responses in rhesus macaques. J Infect Dis.2016;214: Nat Med 1597–1604. 829–838. 15. Sweeney EL, Kallapur SG, Meawad S, et al. Ureaplasma species multi- 34. Divanovic S, Trompette A, Atabani SF, et al. Negative regulation of Toll- ple banded antigen (MBA) variation is associated with the severity of like receptor 4 signaling by the Toll-like receptor homolog RP105. Nat .2005;6:571–578. inflammation in vivo and in vitro in human placentae. Front Cell Infect Immunol Microbiol.2017;7:123. 35. Hoebe K, Du X, Georgel P, et al. Identification of Lps2 as a key transducer of MyD88-independent TIR signalling. Nature.2003;424: 16. O'Neill LA, Sheedy FJ, McCoy CE. MicroRNAs: the fine-tuners of Toll- 743–748. like receptor signalling. Nat Rev Immunol.2011;11:163–175. 17. Vogel SN, Fitzgerald KA, Fenton MJ. TLRs: differential adapter uti- 36. Cha J, Bartos A, Egashira M, et al. Combinatory approaches prevent lization by Toll-like receptors mediates TLR-specific patterns of gene preterm birth profoundly exacerbated by gene-environment interactions. .2013;123:4063–4075. expression. Mol Interv.2003;3:466–477. J Clin Investig 37. Sun X, Cappelletti M, Li Y, Karp CL, Divanovic S, Dey SK. Cnr2 defi- 18. Takeda K, Kaisho T, Akira S. Toll-like receptors. Annu Rev Immunol. 2003;21:335–376. ciency confers resistance to inflammation-induced preterm birth in mice. Endocrinology.2014;155:4006–4014. 19. Koga K, Mor G. Toll-like receptors at the maternal-fetal interface in 38. Robertson SA, Christiaens I, Dorian CL, et al. Interleukin-6 is an essen- normal pregnancy and pregnancy disorders. Am J Reprod Immunol. 2010;63:587–600. tial determinant of on-time parturition in the mouse. Endocrinology. 2010;151:3996–4006. 20. Noguchi T, Sado T, Naruse K, et al. Evidence for activation of Toll-like receptor and receptor for advanced glycation end products in preterm 39. Robertson SA, Skinner RJ, Care AS. Essential role for IL-10 in resistance to lipopolysaccharide-induced preterm labor in mice. . birth. Mediators Inflamm.2010;2010:490406. J Immunol 2006;177:4888–4896. 21. Riley JK, Nelson DM. Toll-like receptors in pregnancy disorders 40. Racicot K, Cardenas I, Wunsche V, et al. Viral infection of the preg- and placental dysfunction. Clin Rev Allergy Immunol.2010;39: 185–193. nant cervix predisposes to ascending bacterial infection. J Immunol. 2013;191:934–941. 22. Thaxton JE, Nevers TA, Sharma S. TLR-mediated preterm birth 41. Romero R, Mazor M, Munoz H, Gomez R, Galasso M, Sherer DM. The in response to pathogenic agents. Infect Dis Obstet Gynecol. 2010;2010:378472. preterm labor syndrome. Ann N Y Acad Sci.1994;734:414–429. CAPPELLETTI ET AL. 543

42. Schenk M, Belisle JT, Modlin RL. TLR2 looks at lipoproteins. Immunity. 59. Trinchieri G. Type I interferon: friend or foe. J Exp Med. 2009;31:847–849. 2010;207:2053–2063. 43. Chandran SS, Verhoeven D, Teijaro JR, Fenton MJ, Farber DL. TLR2 60. Dietrich N, Lienenklaus S, Weiss S, Gekara NO. Murine Toll-like recep- engagement on dendritic cells promotes high frequency effector and tor 2 activation induces type I interferon responses from endolysoso- memory CD4 T cell responses. J Immunol.2009;183:7832–7841. mal compartments. PLoS One.2010;5:e10250. 44. Chua BY, Zeng W, Lau YF, Jackson DC. Comparison of lipopeptide- 61. Barbalat R, Lau L, Locksley RM, Barton GM. Toll-like receptor 2 on based immunocontraceptive vaccines containing different lipid inflammatory monocytes induces type I interferon in response to viral groups. Vaccine.2007;25:92–101. but not bacterial ligands. Nat Immunol.2009;10:1200–1207. 45. Perkins DJ, Polumuri SK, Pennini ME, Lai W, Xie P, Vogel SN. Repro- 62. Racicot K, Kwon JY, Aldo P, et al. TypeI interferon regulates the placen- gramming of murine macrophages through TLR2 confers viral resis- tal inflammatory response to bacteria and is targeted by virus: mech- tance via TRAF3-mediated, enhanced interferon production. PLoS anism of polymicrobial infection-induced preterm birth. Am J Reprod Pathog.2013;9:e1003479. Immunol.2016;75:451–460. 46. Salminen A, Paananen R, Vuolteenaho R, et al. Maternal endotoxin- 63. Buwitt-Beckmann U, Heine H, Wiesmuller KH, Jung G, Brock R, Ulmer induced preterm birth in mice: fetal responses in Toll-like receptors, AJ. Lipopeptide structure determines TLR2 dependent cell activation collectins, and cytokines. Pediatr Res.2008;63:280–286. level. FEBS J.2005;272:6354–6364. 47. Romero R, Miranda J, Chaiworapongsa T, et al. Prevalence and clin- 64. de Weerd NA, Vivian JP, Nguyen TK, et al. Structural basis of a unique ical significance of sterile intra-amniotic inflammation in patients interferon-beta signaling axis mediated via the receptor IFNAR1. Nat with preterm labor and intact membranes. Am J Reprod Immunol. Immunol.2013;14:901–907. 2014;72:458–474. 65. de Weerd NA, Vivian JP, Nguyen TK, et al. Structural basis of a unique 48. Yoshimura K, Hirsch E. Interleukin-6 is neither necessary nor sufficient interferon-beta signaling axis mediated via the receptor IFNAR1. Nat for preterm labor in a murine infection model. J Soc Gynecol Investig. Immunol.2013;14:901–917. 2003;10:423–427. 66. Thaxton JE, Romero R, Sharma S. TLR9 activation coupled to IL- 49. Hanna N, Bonifacio L, Reddy P, et al. IFN-gamma-mediated inhibition 10 deficiency induces adverse pregnancy outcomes. J Immunol. of COX-2 expression in the placenta from term and preterm labor 2009;183:1144–1154. pregnancies. Am J Reprod Immunol.2004;51:311–318. 67. Hamilton SA, Tower CL, Jones RL. Identification of chemokines 50. Zeng W, Ghosh S, Lau YF, Brown LE, Jackson DC. Highly immunogenic associated with the recruitment of decidual leukocytes in human and totally synthetic lipopeptides as self-adjuvanting immunocontra- labour: potential novel targets for preterm labour. PLoS One.2013; ceptive vaccines. J Immunol.2002;169:4905–4912. 8:e56946. 51. Metzger JW, Beck-Sickinger AG, Loleit M, Eckert M, Bessler WG, Jung 68. Malireddi RK, Kanneganti TD. Role of type I interferons in inflam- G. Synthetic S-(2,3-dihydroxypropyl)-cysteinyl peptides derived from masome activation, cell death, and disease during microbial infection. the N-terminus of the cytochrome subunit of the photoreaction centre Front Cell Infect Microbiol.2013;3:77. of Rhodopseudomonas viridis enhance murine splenocyte proliferation. 69. Goritzka M, Durant LR, Pereira C, Salek-Ardakani S, Openshaw PJ, JPeptSci.1995;1:184–190. Johansson C. Alpha/beta interferon receptor signaling amplifies early 52. Muhlradt PF, Kiess M, Meyer H, Sussmuth R, Jung G. Structure proinflammatory cytokine production in the lung during respiratory and specific activity of macrophage-stimulating lipopeptides from syncytial virus infection. JVirol.2014;88:6128–6136. Mycoplasma hyorhinis. Infect Immun.1998;66:4804–4810. 70. Cardenas I, Mor G, Aldo P, et al. Placental viral infection sensitizes 53. Kurokawa K, Kim MS, Ichikawa R, et al. Environment-mediated accu- to endotoxin-induced pre-term labor: a double hit hypothesis. Am J mulation of diacyl lipoproteins over their triacyl counterparts in Reprod Immunol.2011;65:110–117. Staphylococcus aureus. JBacteriol.2012;194:3299–3306. 71. Kwon JY, Romero R, Mor G. New insights into the relationship 54. Gray ML, Singh C. Abortion, stillbirth, early death of young in rab- between viral infection and pregnancy complications. Am J Reprod bits by Listeria monocytogenes.II.Oralexposure.Proc Soc Exp Biol Med. Immunol.2014;71:387–390. 1955;89:169–175. 72. Nilsen NJ, Vladimer GI, Stenvik J, et al. A role for the adaptor pro- 55. Kurokawa K, Ryu KH, Ichikawa R, et al. Novel bacterial lipoproteins TRAM and TRIF in Toll-like receptor 2 signaling. JBiolChem. tein structures conserved in low-GC content Gram-positive bacte- 2015;290:3209–3222. ria are recognized by Toll-like receptor 2. JBiolChem.2012;287: 73. Ng CT, Mendoza JL, Garcia KC, Oldstone MB. Alpha and beta type 13170–13181. 1interferonsignaling:passagefordiversebiologicoutcomes.Cell. 56. Muhlradt PF, Kiess M, Meyer H, Sussmuth R, Jung G. Isolation, struc- 2016;164:349–352. ture elucidation, and synthesis of a macrophage stimulatory lipopeptide from Mycoplasma fermentans acting at picomolar concentration. J Exp Med.1997;185:1951–1958. How to cite this article: Cappelletti M, Lawson MJ, Chan 57. Uchida K, Nakahira K, Mimura K, et al. Effects of Ureaplasma parvum CC, Wilburn AN, Divanovic S. Differential outcomes of lipoprotein multiple-banded antigen on pregnancy outcome in mice. J TLR2 engagement in inflammation-induced preterm birth. Reprod Immunol.2013;100:118–127. JLeukocBiol.2018;103:535–543.https://doi.org/10.1002/ 58. Kang JY, Nan X, Jin MS, et al. Recognition of lipopeptide patterns by Toll-like receptor 2-Toll-like receptor 6 heterodimer. Immunity. JLB.3MA0717-274RR 2009;31:873–884. HEPATOLOGY COMMUNICATIONS, VOL. 2, NO. 5, 2018 Nicotinamide Adenine Dinucleotide Phosphate (Reduced) Oxidase 2 Modulates Inﬂammatory Vigor During Nonalcoholic Fatty Liver Disease Progression in Mice

Rajib Mukherjee,1,2 Maria E. Moreno-Fernandez,1,2 Daniel A. Giles,1-3 Monica Cappelletti,1,2 Traci E. Stankiewicz,1,2 Calvin C. Chan,1-4 and Senad Divanovic1-4

Nonalcoholic fatty liver disease (NAFLD) represents a disease spectrum ranging from benign steatosis to life-threatening cirrhosis and hepatocellular carcinoma. Elevated levels of reactive oxygen species (ROS) and exacerbated inflammatory responses have been implicated in NAFLD progression. Nicotinamide adenine dinucleotide phosphate (reduced) oxidase 2 (NOX2; also known as gp91Phox), the main catalytic subunit of the nicotinamide adenine dinucleotide phosphate (reduced) oxidase complex, modulates ROS production, immune responsiveness, and pathogenesis of obesity-associated metabolic derangements. However, the role of NOX2 in the regulation of immune cell function and inflammatory vigor in NAFLD remains underdefined. Here, we demonstrate that obesogenic diet feeding promoted ROS production by bone marrow, white adipose tissue, and liver immune cells. Genetic ablation of NOX2 impeded immune cell ROS synthesis and was sufficient to uncouple obesity from glucose dysmetabolism and NAFLD pathogenesis. Protection from hepatocellular damage in NOX2-deficient mice correlated with reduced hepatic neutrophil, macrophage, and T-cell infiltration, diminished production of key NAFLD-driving proinflammatory cytokines, and an inherent reduction in T-cell polarization toward Th17 phenotype. Conclusion: Current findings demonstrate a crucial role of the NOX2–ROS axis in immune cell effector function and polarization and consequent NAFLD progression in obesity. Pharmacologic targeting of NOX2 function in immune cells may represent a viable approach for reducing morbidity of obesity-associated NAFLD pathogenesis. (Hepatology Communications 2018;2:546-560)

onalcoholic fatty liver disease (NAFLD) has intestinal microbiome, and exacerbation of proinﬂam- emerged as the most prominent chronic liver matory immune responses.(1,2) Although NAFLD is a N disease. NAFLD, a hepatic manifestation of multifactorial disease, it is well accepted that diet- metabolic disease, represents a wide spectrum of liver induced inﬂammation, reactive oxygen species (ROS) diseases ranging from simple steatosis to nonalcoholic production, and lipid peroxidation play central roles in steatohepatitis (NASH) and cirrhosis. NAFLD pro- NAFLD pathogenesis.(3-5) gression involves a complex interplay between various ROS, the general byproduct of cellular metabolism, biological processes, including obesity, dysbiosis of the impacts macromolecular integrity, cell signaling, and

Abbreviations: ALT, alanine aminotransferase; BAT, brown adipose tissue; BMDC, bone marrow derived dendritic cell; BMDM, bone marrow derived macrophage; BMN, bone marrow neutrophil; CD, clusters of differentiation; DCFDA, 20,70-dichloroﬂuorescin diacetate; eWAT, epididymal white adipose tissue; FBS, fetal bovine serum; HFD, high-fat diet; IFN-c, interferon-c; IL, interleukin; iWAT, inguinal white adipose tissue; LPS, lipopolysaccharide; NAFLD, nonalcoholic fatty liver disease; NASH, nonalcoholic steatohepatitis; NOX, reduced nicotinamide adenine dinucleotide phosphate oxidase; ROS, reactive oxygen species; RPMI, Roswell Park Memorial Institute; TG, triglyceride; Th, T helper; TLR, toll-like receptor; TNF-a, tumor necrosis factor a; UCP1, uncoupling protein 1; WT, wild type. Received October 20, 2017; accepted February 2, 2018. Additional Supporting Information may be found at onlinelibrary.wiley.com/doi/10.1002/hep4.1162/full.

546 HEPATOLOGY COMMUNICATIONS, Vol. 2, No. 5, 2018 MUKHERJEE ET AL. innate immunity. Cell intrinsic imbalance between Specifically, we show that despite similar obesity, adi- ROS production and antioxidant capability results in posity, and liver triglyceride (TG) accumulation, oxidative stress, while oxidative stress contributes to NOX2-deficient mice were protected from obesity- the pathogenesis of various metabolic diseases.(6) In induced glucose dysmetabolism and hepatocellular obesity, excess systemic nutrients induce oxidative damage. Reduced severity of obesity-driven hepatocel- stress and promote physiologic changes in key periph- lular damage in NOX2-deficient mice was associated eral tissues including adipose tissue, liver, and skeletal with decreased hepatic infiltration of immune cells. muscle. The nicotinamide adenine dinucleotide phos- Mechanistically, NOX2-deficient hepatic neutrophils, phate (reduced) oxidase (NOX) enzyme complex, a macrophages, and T cells had diminished production central source of intracellular ROS, is composed of of key NAFLD-driving, proinflammatory cytokines, NOX2 (also known as gp91Phox); the cytosolic including interleukin (IL)-17A, interferon-c (IFN-c), p40Phox, p47Phox, p67Phox units; and Rac family small and tumor necrosis factor a (TNF-a). Further, T-cell guanosine triphosphatase 1 complexes.(7) NOX2, intrinsic NOX2 expression was sufficient to modulate which is highly expressed in immune cells,(8) modu- polarization and activation-driven proinflammatory lates insulin resistance, adipose tissue inflammation, cytokine production in T cells. In sum, our data sug- hepatic stellate cell activation, and hepatic fibrosis.(9-11) gest that NOX2-driven modulation of immune cell However, the role of NOX2 in modulating immune inflammatory capacity may play important roles in cell responsiveness in NAFLD has not been examined. obesity-driven NAFLD progression. A better understanding of mechanisms underlying activation of NOX2-dependent ROS production may uncover unappreciated molecular targets for novel Materials and Methods NAFLD therapies. MICE Here, we demonstrated that high-fat diet (HFD) feeding promoted immune cell ROS generation and All studies were approved by the Cincinnati Child- that NOX2 modulated ROS synthesis, immune cell ren’s Hospital Medical Center’s Institutional Animal inflammatory capacity, and obesity pathogenesis. Care and Use Committee. All mice used in this study

Supported in part by National Institutes of Health awards R01DK099222 (to S.D.), T32AI118697 (to D.A.G. and C.C.C), and P30 DK078392 Pathology of the Digestive Disease Research Core Center at the Cincinnati Children’s Hospital Medical Center; Cincinnati Children’s Hospital Medical Center Pediatric Diabetes and Obesity Center (to S.D.); and American Heart Association 17POST33650045 (to M.E.M.F). Present address for Daniel A. Giles is La Jolla Institute for Allergy and Immunology, La Jolla, CA. Present address for Monica Cappelletti is Divisions of Neonatology and Developmental Biology, David Geffen School of Medicine at University of California, Los Angeles, Mattel Children’s Hospital, Los Angeles, CA. Copyright VC 2018 The Authors. Hepatology Communications published by Wiley Periodicals, Inc., on behalf of the American Association for the Study of Liver Diseases. This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made. View this article online at wileyonlinelibrary.com. DOI 10.1002/hep4.1162 Potential conflict of interest: Nothing to report. ARTICLE INFORMATION:

From the 1Department of Pediatrics, University of Cincinnati College of Medicine, Cincinnati, OH 45220; 2Division of Immunobiology, Cincin- nati Children’s Hospital Medical Center, Cincinnati, OH 45229, USA; 3Immunology Graduate Program and 4Medical Scientist Training Pro- gram, Cincinnati Children’s Hospital Medical Center and the University of Cincinnati College of Medicine, Cincinnati, OH 45220, USA. ADDRESS CORRESPONDENCE AND REPRINT REQUESTS TO:

Senad Divanovic, Ph.D. 3333 Burnet Avenue Division of Immunobiology, Cincinnati Children’s Hospital Cincinnati, OH 45229-3039 Medical Center E-mail: [email protected] TCHRF - Location S, Room #S.5.409 Tel: 11-513-636-0286

547 MUKHERJEE ET AL. HEPATOLOGY COMMUNICATIONS, May 2018 were on a C57BL/6 background, housed in the same using a 100-lm nylon cell strainer, centrifuged at room within a specific pathogen-free facility but not as 1,000g for 10 minutes, treated with erythrocyte lysis littermates, fed food and water ad libitum, and pro- buffer (Lonza), and washed in RPMI medium prior to vided care in accordance with the Guide for the Care ROS and cytokine-level quantification. Liver tissue and Use of Laboratory Animals. Wild-type (WT) samples were minced using gentleMACS tubes (Mil- mice (strain name C57BL/6J) and NOX22/2 mice tenyi Biotec) according to the manufacturer’s recom- (strain name B6.129S-CybbtmDin/J) were obtained mendation. Immune cells were purified using 33% from Jackson Laboratories and were subsequently bred Percoll (Sigma-Aldrich) in RPMI medium followed at the Cincinnati Children’s Hospital Medical Center. by centrifugation at 250g for 20 minutes at room temperature and then treated with erythrocyte lysis buffer OBESITY AND NAFLD MODEL and washed in RPMI medium prior to ROS and cytokine-level quantification. Starting at 8 weeks of age, mice were fed ad libitum either chow diet (Lab diet #5010) or HFD (Research diets #D12492) for 8 or 22 weeks, with fresh food pro- ROS STAINING vided on a weekly basis. Body weight and food con- ROS production was quantified ex vivo by 20,70- sumption were monitored weekly. dichlorofluorescin diacetate (DCFDA) staining. Iso- lated immune cells were incubated with 25 lM IMMUNE CELL ISOLATION AND DCFDA at 37oC in complete RPMI medium for 30 CYTOKINE PRODUCTION minutes, washed with phosphate-buffered saline, and analyzed by flow cytometry. For each experiment, at Peritoneal neutrophils (PN) were collected by peri- least 1 3 105 cells/mouse were analyzed to calculate toneal gavage after intraperitoneal injection of zymosan mean fluorescence intensity of DCFDA staining. (10, 100, or 1,000 lg/mouse; 12 hours) as (12) described. Bone marrow-derived macrophages SERUM PARAMETERS (BMDMs), bone marrow neutrophils (BMNs), and bone marrow-derived dendritic cells (BMDCs) were TG quantification was performed as reported.(14,17-19) isolated as described.(12-15) Briefly, bone marrow cells Briefly, 200 lLoftriglyceridereagent(PointeScientific) were differentiated into BMDMs or BMDCs using was added to 10 lLofserumina96-well,clear,flat- Roswell Park Memorial Institute (RPMI) medium bottom plate (Costar). Standards (Pointe Scientific) were (Gibco) supplemented with 10% fetal bovine serum prepared according to the manufacturer’s instructions. (FBS), 1% penicillin/streptomycin, 1% glutamine, and Serum leptin (Millipore) and NO2 (Sigma-Aldrich) lev- 20 ng/mL macrophage colony stimulating factor or els were quantified using commercially available enzyme- granulocyte-macrophage colony stimulating factor linked immunosorbent assay kits. o (Biolegend) for 6 days at 37 C and 5% CO2.BMNs were purified using the anti-Ly-6G Microbead Kit HISTOLOGY (Miltenyi Biotec) and cultured in RPMI medium supplemented with 10% FBS, 1% penicillin/streptomycin, Liver and eWAT samples were fixed in 10% forma- and 1% glutamine. BMDMs (6 3 105 cells/mL), lin. Hematoxylin and eosin staining was performed on BMDCs (6 3 105 cells/mL), and BMNs (5 3 105 5-lm sections from the paraffin-embedded tissue blocks cells/mL) were stimulated with lipopolysaccharide for conventional light microscopy analysis. Terminal (LPS; 10, 100, or 1,000 ng/mL) for 4 hours, followed deoxynucleotidyl transferase–mediated deoxyuridine tri- by ROS quantification and collection of cell superna- phosphate nick-end labeling staining of eWAT samples tant for cytokine analysis by enzyme-linked immuno- were deparaffinized and stained using the standard pro- sorbent assay (BD Opt EIA; BD Bioscience). For tocol. For immunohistochemical staining, liver sections tissue immune cell isolation, single-cell suspensions were deparaffinized and immunostained with rabbit from epididymal white adipose tissue (eWAT) and anti-mouse clusters of differentiation (CD)68 antibody liver were obtained using common methods.(14,16,17) (Abcam) or rabbit monoclonal anti-mouse CD3 anti- Minced eWAT tissue was digested for 45 minutes in body (Ventana Medical Systems, Inc.), using the auto- collagenase cocktail (0.03 mg/mL liberase and 50 U/ mated Ventana immunostainer according to the mL deoxyribonuclease I [Sigma-Aldrich]), filtered manufacturer’s recommendation.(14,17,18)

548 HEPATOLOGY COMMUNICATIONS, Vol. 2, No. 5, 2018 MUKHERJEE ET AL.

GLUCOSE AND INSULIN (Zombie UV Dye: Biolegend) and with directly conju- SENSITIVITY TESTS gated monoclonal antibodies to CD45-PEDazzle (clone 104), CD4-allophycocyanin (APC)ef780 (clone All studies were performed on mice fasted overnight RM4-5), CD3-Alexa Fluor 700 (clone 17A2), (14,17) as reported. Briefly, for the glucose tolerance test, CD11b-ef450 (clone M1/70), CD11c-APC (clone mice were challenged with a bolus of dextrose (100 N418), Gr1- fluorescein isothiocyanate (FITC) (clone mg/kg body weight) by intraperitoneal injection and RB6-8C5), and B220-BV605 (clone RA3-6B2) in blood glucose levels were measured at 20, 40, 60, 90, phosphate-buffered saline supplemented with 2% FBS and 120 minutes. For the insulin tolerance test, mice for 30 minutes. Cells were subsequently fixed, permea- were challenged with insulin (0.3 U/mL of insulin/10 bilized, and stained with IFN-c–phycoerythrin (PE)– g of body weight) by intraperitoneal injection and Cy7 (clone XMG1.2), TNF-a–BV650 (clone MP6- blood glucose levels were measured at the time points XT22), and IL-17A–Percp Cy5.5 (clone eBio17B7) specified above. All measurements were performed (all antibodies from e-Bioscience) for 30 minutes. LSR using the Accu-chek kit (Roche). Fortessa (BD Biosciences) and FlowJo software (version X0.7) were used for flow cytometry data collection LIVER TGS and analysis.(14,17,18) Hepatic TG levels were quantified as described.(14,17,18) Briefly, standards and samples were T-HELPER 17 POLARIZATION added to a 96-well, clear, flat-bottom plate (Costar) AND T-CELL ACTIVATION containing 200 lL of triglyceride reagent (Pointe Sci- T-cell isolation kits (Miltenyi Biotec) were used to entific). Hepatic TGs were quantified at 500-520 nm isolate naive splenic T cells according to the manufac- using the vmax Microplate Reader (Molecular Devi- turer’s protocol. Isolated CD41CD62L1 T cells were ces) and SoftMax Pro version 5 software. cultured in a goat anti-hamster immunoglobulin G- coated plate and differentiated for 6 days with anti- HEPATOCELLULAR DAMAGE CD3 (1 lg/mL) and anti-CD28 (0.5 lg/mL) and Alanine aminotransferase (ALT) levels were quanti- with or without recombinant human IL-1b (1 ng/ fied as described.(14,18) Briefly, 10 lL of mouse serum mL), recombinant mouse IL-6 (10 ng/mL), recombi- was added to a 96-well flat-bottom plate (Costar) con- nant human IL-23 (20 ng/mL), or recombinant taining 200 lL ALT buffer that was prepared by mix- human TGF-b1 (1 ng/mL). T cells were stimulated ing ALT Activator Reagent and ALT sample Diluent using phorbol 12-myristate 13-acetate (50 ng/mL), Reagent (Catachem Inc.). Catatrol I and II (Catachem ionomycin (1 lg/mL), and brefeldin A (10 lg/mL) for Inc.) were used as controls and were prepared accord- 4 hours, and IL-17A, IFN-c, and TNF-a production ing to the manufacturer’s instructions. ALT levels were quantified by flow cytometry. For each experi- were quantified using the BioTek Synergy 2 Multi- ment, 1 3 105 to 2 3 105 cells per mouse were Mode Microplate Reader with Gen5 version 2 analyzed. software. PERIPHERAL BLOOD CELLULAR FLOW CYTOMETRY COMPOSITION Immune cells were stained directly with conjugated Blood samples were collected in heparin tubes (BD monoclonal antibodies or isotype controls (all eBio- Biosciences) and analyzed in Hemavet 950 (Drew science). For quantification of immune cell cytokine Scientific Inc.) using the manufacturer’s program for production, isolated immune cells were stimulated for mouse blood analysis. 4 hours with 50 ng/mL phorbol 12-myristate 13- acetate (Sigma-Aldrich) and 1 lg/mL ionomycin STATISTICAL ANALYSIS (Calbiochem), in the presence of brefeldin A (10 lg/mL; Sigma-Aldrich) and analyzed by intracellular flow Data were analyzed using the unpaired Student t cytometry. Briefly, cells were treated with fluorescence- test. All values are represented as means 1 SE and activated cell sorting buffer (BD Biosciences) for 15 graphed with Prism 7 software (GraphPad Software minutes, washed, and stained with live/dead stain Inc.).

549 MUKHERJEE ET AL. HEPATOLOGY COMMUNICATIONS, May 2018

compared to the chow diet significantly enhanced Results ROS production by primary BMNs (Fig. 1A). How- OBESITY AND INFLAMMATION, ever, HFD feeding failed to promote ROS production IN PART BY NOX2 ACTIVATION, in primary BMDMs and BMDCs, both being immune cells derived by prolonged in vitro culture and PROMOTE IMMUNE CELL ROS polarization (Fig. 1A; Supporting Fig. S1). Further, PRODUCTION HFD feeding significantly enhanced ROS production 1 high Obesity-associated inflammation and oxidative by neutrophils (CD11b Gr1 ) and macrophages 1 1 stress, in part by ROS production, modulate the path- (CD11b F4/80 ) infiltrating eWAT and liver (Fig. ogenesis of obesity-associated sequelae, including type 1B,C), which are tissues relevant to NAFLD patho- 2 diabetes and NAFLD.(1,2,4,5,20) Here, we examined genesis. Genetic ablation of NOX2 was sufficient to the contribution of obesogenic diet feeding on immune impede obesity-driven ROS induction in BMNs and cell ROS production in various tissues. HFD feeding macrophages in the context of inflammatory challenge

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

FIG. 1. HFD modulates ROS production by immune cells. (A) ROS production by bone marrow neutrophils and macrophages from either chow diet or HFD-fed WT mice. (B) ROS production by hepatic neutrophils and macrophages from either chow- or HFD- fed WT mice. (C) ROS production by epididymal white adipose tissue neutrophils and macrophages from either chow- or HFD-fed WT mice. (D) ROS production by bone marrow neutrophils and macrophages from HFD-fed WT and NOX22/2 mice. (A-D) Data presented as relative mean ﬂuorescence intensity. Gray bars denote WT chow-fed mice; white bars denote WT HFD-fed mice; black bars denote NOX22/2 HFD-fed mice. (A,D) A single experiment, n 5 3-4 mice/condition. (B,C) Data combined from three independent experiments, n 5 6-8 mice/condition. Data represent means 1 SE. Student t test, *P < 0.05, **P < 0.01, ***P < 0.001. Abbreviations: CD, chow diet; MFI, mean ﬂuorescence intensity.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

550 HEPATOLOGY COMMUNICATIONS, Vol. 2, No. 5, 2018 MUKHERJEE ET AL.

(Fig. 1D). Obesity is associated with metabolic endo- increased inguinal white adipose tissue [iWAT]), toxemia(17,21) and various inflammatory triggers, increased adipose tissue inflammation, and adipocyte including endotoxin (LPS), that drive ROS and proin- death have been correlated with advancement of flammatory cytokine production.(2,4) NOX2 is a cen- obesity-related sequelae, including NAFLD.(24) Nota- tral player in immune cell ROS synthesis. Hence, the bly, obese NOX22/2 mice compared to WT controls impact of NOX2 on immune cell ROS production in exhibited increased eWAT weight and decreased response to endotoxin sensing was examined next. iWAT weight (Fig. 2G). The difference in eWAT NOX2-deficient neutrophils, BMDMs, and BMDCs mass was independent of adipocyte size (Fig. 2H) but compared to WT controls exhibited reduction in LPS- correlated with decreased eWAT adipocyte cell death induced ROS production in a dose-dependent manner as determined by terminal deoxynucleotidyl transfer- (Supporting Fig. S2). Further, LPS-induced ROS lev- ase–mediated deoxyuridine triphosphate nick-end els were higher in BMNs isolated from HFD-fed mice labeling staining (Fig. 2I). compared to chow-fed controls (data not shown). The role of NOX2-dependent ROS in modulating These effects were not limited to LPS as zymosan the severity of obesity-associated glucose dysmetabo- treatment of NOX2-deficient mice resulted in similarly lism is controversial.(9,10,25,26) Hence, the development decreased neutrophil ROS production (Supporting of glucose dysmetabolism and insulin resistance was Fig. S3A). Further, the NOX2 impact on zymosan- examined. Although obese NOX22/2 mice compared induced ROS production was independent of differen- to WT controls had reduced fasting glucose levels and tial neutrophil recruitment into the peritoneal cavity improved glucose tolerance (Fig. 3A,B), lack of NOX2 (Supporting Fig. S3B). Congruently, stimulation of did not alter systemic insulin levels or insulin sensitiv- NOX2-deficient macrophages with LPS resulted in ity (Fig. 3C,D). Further, genetic ablation of NOX2 reduced IL-6 production (data not shown), a proin- did not modulate the expression of mediators central flammatory mediator known to play a role in patho- for insulin-induced glucose homeostasis in liver and genesis of obesity-associated sequelae.(22) Collectively, eWAT (e.g., glucose transporter 1 [Glut1], Glut4, these data suggest that obesity augments immune cell carbohydrate-responsive element-binding protein b, ROS production in tissues that contribute to the path- and sterol regulatory element binding protein 1c; data ogenesis of obesity-associated sequelae. Further, our not shown).(27,28) Overall, these data suggest that data suggest that NOX2, in the context of HFD- NOX2 regulates glucose but not insulin metabolism in induced obesity or an obesity-associated inflammatory HFD-induced obesity. Of note, these data are in environment, likely promotes immune cell-driven agreement with reported NOX2-independent path- ROS synthesis. ways of glucotoxicity in pancreatic islets in mice.(29)

NOX2 UNCOUPLES OBESITY AND NOX2 REGULATES HFD-INDUCED ADIPOSITY FROM GLUCOSE HEPATOCELLULAR DAMAGE DYSMETABOLISM AND NAFLD PROGRESSION As NOX2 modulated immune cell ROS and proin- Obesity and glucose dysmetabolism contribute to ﬂammatory cytokine production with HFD feeding, NAFLD pathogenesis. Thus, we examined the impact the impact of NOX2 in obesity development and path- of NOX2 on NAFLD development. HFD-fed WT ogenesis of obesity-associated sequelae was examined. and NOX22/2 mice exhibited similar liver weight In agreement with a published report,(11) both WT (Fig. 4A) and hepatic TG accumulation (Fig. 4B). and NOX22/2 mice exhibited similar HFD-driven However, in contrast to obese WT controls, which had weight gains (Fig. 2A). Comparable weight gain corre- macrovesicular steatosis (a characteristic that is highly lated with similar food intake, systemic TG, leptin and prevalent in most patients with NAFLD), obese 2/2 NO2 levels, brown adipose tissue (BAT) mass, and NOX2 mice primarily exhibited microvesicular BAT uncoupling protein 1 (Ucp1) expression (Fig. steatosis (Fig. 4C,D).(30,31) Attenuated macrovesicular 2B-F). Notably, analogous BAT characteristics are steatosis observed in obese NOX22/2 mice correlated suggestive of similar overall energy expenditure with reduced serum ALT levels (Fig. 4E) and hepatic between WT and NOX22/2 mice and are in agree- lipid peroxidation (Fig. 4F). Of note, elevated hepatic ment with published ﬁndings.(23) Altered adipose tis- TG and ROS levels are key characteristics of macro- sue mass distribution (e.g., decreased eWAT and steatotic livers, something that renders hepatocytes

551 MUKHERJEE ET AL. HEPATOLOGY COMMUNICATIONS, May 2018

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

FIG. 2. NOX2 modulates HFD-driven white adipose tissue distribution. Eight-week-old WT and NOX22/2 mice were fed an HFD for 22 weeks. (A) Body weight; (B) cumulative weekly food intake; (C) serum leptin level; (D) serum triglyceride level; (E) serum nitrite level; (F) BAT weight and Ucp1 expression; (G) WAT weight; (H) hematoxylin and eosin staining of eWAT; (I) TUNEL staining of eWAT and TUNEL-positive cell number per ﬁeld of view. White bars/squares denote WT mice; black bars/ squares denote NOX22/2 mice. (A-I) A representative of two individual experiments, n 5 4-6 mice/condition. Data represent means 1 SE. Student t test, *P < 0.05, **P < 0.01. Abbreviations: AU, arbitrary unit (relative); TUNEL, terminal deoxynucleotidyl transferase–mediated deoxyuridine triphosphate nick-end labeling.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! susceptible to secondary insults and cellular death.(32) in HFD-fed WT and NOX22/2 mice was examined. Such findings are in agreement with the notion that, A reduction in total hepatic immune cell infiltration along with mitochondrial ROS, NOX2-dependent (CD451 cells) observed in obese NOX22/2 mice chronic oxidative stress and ROS-induced lipid peroxi- largely correlated with robustly lowered numbers of dation are central to NAFLD progression.(33) liver-infiltrating macrophages (CD11b1Gr1Low) and While biological processes responsible for progres- CD41 T cells (Fig. 4G), a finding further confirmed sion of NAFL to NASH are not fully understood, by a significant reduction in CD68 and a trend toward augmented infiltration of immune cells and enhanced reduced CD3 immunostaining of liver sections (Fig. production of proinflammatory mediators by liver- 4H,I). Of note, altered hepatic immune infiltration infiltrating immune cells are believed to play an impor- observed in obese NOX22/2 mice was independent of tant role.(2,4) Hence, hepatic immune cell composition differential circulating immune cell numbers or

552 HEPATOLOGY COMMUNICATIONS, Vol. 2, No. 5, 2018 MUKHERJEE ET AL.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

FIG. 3. NOX2 regulates HFD-induced glucose dysmetabolism. Eight-week-old WT and NOX22/2 mice were fed an HFD for 22 weeks. (A) Fasting glucose; (B) glucose tolerance test after 13 weeks of HFD feeding; (C) fasting insulin; (D) insulin tolerance test performed after 15 weeks of HFD feeding. White bars/squares denote WT mice; black bars/squares denote NOX22/2 mice. (A-D) A representative of two individual experiments, n 5 4-6 mice/condition. Data represent means 1 SE. Student t test, *P < 0.05, **P < 0.01. Abbreviations: AUC, area under the curve; GTT, glucose tolerance test; ITT, insulin tolerance test.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! composition in the context of HFD feeding or follow- by liver-infiltrating neutrophils (CD11b1Gr1high; Fig ing systemic LPS challenge (Supporting Fig. S4). 5A), macrophages (CD11b1Gr1low; Fig 5B), and IL-17A, IFN-c, and TNF-a by CD41 T cells (CD31CD41; Fig. 5C). Together, these data suggest NOX2 MODULATES HEPATIC that NOX2 modulates inflammatory vigor in liver- IMMUNE CELL INFLAMMATORY infiltrating immune cells. VIGOR Although NOX2 is predominantly associated with myeloid immune cell function, the impact of NOX2 Hepatic accumulation of neutrophils, macrophages, on T-cell cytokine production and whether such 1 and CD4 T cells (e.g., T helper [Th]1 and Th17 effects are T-cell intrinsic are unknown. Th17 cells (34,35) subsets) correlates with NAFLD progression. are the primary producers of IL-17A, a proinflamma- NOX2 regulates hepatic immune cell infiltration and tory cytokine known to propagate NAFLD patho- (36) immune cell cytokine production. Notably, proin- genesis.(17) Thus, we examined the role of NOX2 on flammatory cytokines (e.g., IFN-c,TNF-a, IL-17A) Th17 cell polarization. Compared to WT cells, are well-established modulators of NAFLD pathogen- NOX22/2 primary naive splenic CD41 Tcellscul- esis.(1,2,4,20) We therefore examined the role of NOX2 tured under Th17 polarizing conditions exhibited in the modulation of immune cell inflammatory vigor. reduced Th17 polarization and IL-17A production Obese NOX22/2 mice compared to WT controls (Fig. 6A). The effects on cytokine production were exhibited attenuated production of IFN-c and TNF-a not unique to Th17 cells as primary naive splenic

553 MUKHERJEE ET AL. HEPATOLOGY COMMUNICATIONS, May 2018

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

FIG. 4. NOX2 regulates HFD-induced NAFLD pathogenesis. Eight-week-old WT and NOX22/2 mice were fed an HFD for 22 weeks. (A) Liver weight; (B) liver triglyceride content; (C) hematoxylin and eosin staining of liver; (D) Oil-Red-O staining of liver; (E) ALT levels; (F) liver 4-HNE levels. (G) Gating strategy for quantification of hepatic immune cell infiltration by flow cytometry and hepatic immune cell composition. CD11b1 Gr1low macrophage populations were further gated by forward side scatter to differentiate between CD11b1Gr1low macrophages and leukocytes. (H) Liver CD31 staining; (I) liver CD681 staining. White bars denote WT mice; black bars denote NOX22/2 mice. (A-I) A representative of two individual experiments, n 5 4-6 mice/condition. Data represent means 1 SE. Student t test, *P < 0.05, **P < 0.01. Abbreviations: 4-HNE, 4-hydroxynonenal; FSC, Forward scatter; SSC, Side scatter.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

CD41 TcellsisolatedfromNOX22/2 mice were less NOX2 plays an intrinsic role in T-cell responsiveness poised to produce IFN-c and TNF-a following poly- and Th17 cell polarization. Importantly, such effects clonal T-cell receptor stimulation using antiCD3/ may confer modulation of the hepatic inﬂammatory antiCD28 (Fig. 6B). In sum, these data suggest that milieu in NAFLD progression.

554 HEPATOLOGY COMMUNICATIONS, Vol. 2, No. 5, 2018 MUKHERJEE ET AL.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

FIG. 5. NOX2 regulates hepatic immune cell inflammatory vigor. Eight-week-old WT and NOX22/2 mice were fed an HFD for 22 weeks, and cytokine production in liver infiltrating immune cells was quantified by flow cytometry. Cells were exclusively identified from neutrophil, macrophage, and CD31CD41 gates described in Fig. 4G. (A) Cytokine production by liver-infiltrating neutrophils (CD11b1GR1high). (B) Cytokine production by liver-infiltrating macrophages (CD11b1GR1low). (C) Cytokine production by liver- infiltrating T cells (CD31CD41). White bars denote WT mice; black bars denote NOX22/2 mice. (A-C) A representative of two individual experiments, n 5 4-6 mice/condition. Data represent means 1 SE. Student t test, *P < 0.05, **P < 0.01.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! Discussion and type 2 diabetes mellitus.(20) Despite the clinical significance, the underlying mechanisms regulating NAFLD, the most prominent liver disease world- NAFLD development and pathogenesis remain wide, is directly linked with obesity, insulin resistance, underdefined and represent a significant gap in

555 MUKHERJEE ET AL. HEPATOLOGY COMMUNICATIONS, May 2018

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

FIG. 6. NOX2 modulates T-cell polarization and effector function. Eight-week-old WT and NOX22/2 mice were fed an HFD for 8 weeks and used for analysis of T-cell responses. (A) Representative ﬂow plot of naive splenic CD41 T cells toward Th17 differentiation and percentage of IL-17A1 and IFN-c1 in polarized Th17 cells. (B) Representative ﬂow plot of CD3/CD28-stimulated naive CD41 T cells and percentage of TNF-a1 and IFN-c1 production by CD41 T cells. White bars denote WT mice; black bars denote NOX22/2 mice. (A,B) A single experiment, n 5 3 mice/condition. Data represent means 1 SE. Student t test, **P < 0.01.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! knowledge. It is well appreciated that oxidative stress NOX2 modulates neutrophil and macrophage ROS and subsequent alterations to inflammatory cascades expression in WAT and liver has not been determined. contribute to the pathogenesis of obesity-associated Future studies focused on the role of NOX2 in ROS sequelae,(37) but whether NOX2 regulates NAFLD synthesis, proinflammatory cytokine production, neu- pathogenesis and NAFLD-relevant immune cell trophil recruitment, and macrophage polarization in inflammatory vigor has not been determined. Our data obesity/NAFLD-relevant tissues are clearly warranted. provide novel insights about the role of NOX2 in obe- Notably, obesity is associated with metabolic endotox- sity and obesity-associated NAFLD pathogenesis. emia,(2,21) and TLR4 signaling plays an important role Specifically, here we show that in the context of HFD- in NAFLD pathogenesis.(2,14,38) Hence, the interplay induced obesity, NOX2 modulates (a) immune cell between TLR4 and NOX2 activation and ROS pro- ROS production, (b) immune cell inflammatory capac- duction in obesity warrants further investigation. ity pivotal for NAFLD progression, (c) glucose dysme- Our data also suggest that the NOX2 effects on tabolism, and (d) development of steatohepatitis and immune cell ROS production are not endotoxin and/ hepatocellular damage. or TLR4-signaling specific. Along with TLR4, TLR2 Under obesogenic stress, various immune cells pro- is a key contributor of development of NASH through duce elevated levels of ROS in bone marrow, WAT, inflammasome activation.(39) Notably, neutrophil- and liver. Genetic ablation of NOX2 correlated with driven ROS production in the context of zymosan attenuated ROS generation and proinflammatory cyto- sensing, a component of fungal cell walls and an acti- kine production in bone marrow immune cells follow- vator of TLR2 signaling, induced similar outcomes. ing HFD feeding and endotoxin-driven activation of Although zymosan levels are not commonly quantified toll-like receptor (TLR)4 signaling. However, whether in obesity, intestinal colonization with fungi has been

556 HEPATOLOGY COMMUNICATIONS, Vol. 2, No. 5, 2018 MUKHERJEE ET AL. implicated in NAFLD pathogenesis.(40,41) Whether associated disease progression.(24) An inverse propor- other TLRs or innate immune receptors regulate acti- tion of eWAT to iWAT size coupled with an vation of the NOX2–ROS axis and the contribution of increased presence of apoptotic adipocytes in eWAT such activation in NAFLD pathogenesis are unknown adipocytes were supportive of advanced adiposity in and should be examined. HFD-fed WT mice compared to NOX22/2 mice.(24) Experimental and clinical evidence suggest a com- Our findings suggest that NOX2 does not modulate plex interplay between TLRs, metabolic endotoxemia, HFD-induced body weight gain or thermogenic and intestinal microbiome in NAFLD progres- capacity but plays an important role in WAT redistri- sion.(1,2,14,21,41) Interactions between the microbiome, bution and apoptosis. As genetic ablation of NOX2 alcohol-producing bacteria, ROS, and the immune modulates WAT distribution, adipocyte death, and system are believed to play an important role in NASH the type of hepatic steatosis, a better understanding of progression.(42) Hence, future investigations employing NOX2 contribution in WAT physiology, adipocyte littermate controls in the analysis of the NOX2–ROS function, and lipid partitioning is required. axis in obesity-associated alterations of the intestinal Glucose dysmetabolism, a consequence of excessive microbiome and its contributions to NAFLD develop- systemic nutrient load, WAT redistribution and ment and progression are of obvious importance. inflammation, and imbalance in the ROS axis, repre- NOX2-mediated effects on ROS production were sents one of the most common end-organ sequelae of not specific to a single immune cell type (e.g., macro- diet-induced obesity.(33) Our data suggest that despite phages, neutrophils), suggesting that NOX2 broadly similar obesity, systemic insulin production, and insu- affects immune responsiveness and, as such, likely lin tolerance in NOX22/2 mice compared to WT con- contributes to the pathogenesis of a variety of dis- trols, NOX22/2 mice were protected from HFD- eases, including obesity. Specifically, our data demon- induced glucose dysmetabolism. These findings sug- strated that prolonged culture of primary cells outside gest that divergent mechanisms,(45) NOX2-dependent the obesogenic environment, as seen with BMDMs and independent, are likely to impact glucose metabo- and BMDCs, resulted in similar baseline ROS levels lism and insulin sensitivity in HFD-fed mice. Further, in NOX22/2 and WT cells. Challenge with an the divergence between NOX2 deficiency and insulin inflammatory stimulant was sufficient to uncover the sensitivity may have connections with a largely under- differences in ROS production in such cells. These appreciated role of ROS in stimulating tyrosine data suggest that the obesity-associated inflammatory phosphorylation-dependent insulin signaling.(46) In environment plays an important role in NOX2 activa- fact, the observed impact on glucose dysmetabolism, a tion. However, specific pathways and cellular mecha- well-known attribute of patients with NAFLD, war- nisms directly associated with NOX2-dependent rants a further, detailed, mechanistic analysis on the effects in disease pathogenesis remain underdefined. role of NOX2 in glucose metabolisms and its interplay Our data demonstrated that diminished ROS pro- with NAFLD development. duction in NOX22/2 mice did not affect HFD- NOX2 is highly expressed in the liver, and a serum- induced obesity, adiposity, and food consumption. soluble NOX2-derived peptide (sNOX2-dp) is a known Thermogenic capacity, which in part is dependent on marker of liver steatosis.(47) However, how NOX2 regu- BAT size and UCP-1 expression/function, is central to lates immune responses in NAFLD progression is prin- whole body metabolism in both mice and humans.(43) cipally underdefined. Here, we demonstrated that Furthermore, augmented mitochondrial ROS produc- NOX22/2 mice are protected from HFD-induced tion represents a key step in Ucp1 activation and ther- macrosteatosis, lipid peroxidation, and hepatocellular mogenesis.(44) In agreement with similar total body damage despite similar liver weights and TG content. weight gain between obese WT and NOX22/2 mice, The reduced disease severity of hepatocellular damage NOX22/2 mice exhibited similar BAT weight and correlated with attenuated hepatic immune cell infiltra- Ucp1 expression. These findings suggest that NOX2- tion, including macrophages, neutrophils, and T cells. dependent ROS does not regulate the thermogenic state Infact, Diet-induced hepatic neutrophil, macrophage, in an experimental model of HFD-induced obesity and and T-cell infiltration are trademarks of NAFLD pro- that such effects are likely dependent on other, and pos- gression and pathogenesis.(1,35,48) The reduction in sibly redundant, mechanisms of ROS induction. hepatic macrophages in livers of NOX22/2 mice is con- Additionally, WAT redistribution and adipocyte sistent with a recent report of attenuated macrophage death during weight gain is associated with obesity- infiltration into WAT in myeloid-specific NOX22/2

557 MUKHERJEE ET AL. HEPATOLOGY COMMUNICATIONS, May 2018

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

FIG. 7. Summary figure depicting immunomodulatory roles of NOX2 in HFD-induced NAFLD progression. NOX2 deficiency correlates with protection from HFD-induced glucose insensitivity, hepatocellular damage, and hepatic immune cell infiltration, despite similar total body weight gain and hepatic steatosis. Mechanistically, protection from the pathogenesis of obesity-associated sequelae correlated with NOX2-driven modulation of ROS production and immune cell inflammatory capacity (e.g., macrophages, CD41 T cells, and Th17 polarization). These data suggest a previously unappreciated role of the NOX2–ROS axis in the interplay between inflammatory vigor during NAFLD pathogenesis.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! mice.(36) Altered number of liver-infiltrating immune Further confirmation of the immune NOX2 expression cells in NOX22/2 mice also suggest reduced activation as the central locus of disease pathology, by use of bone of inflammatory axes in these mice. Proinflammatory marrow chimera studies or immune cell-specific NOX2 cytokine production, by liver-infiltrating immune cells, deletion, should be performed in the context of obeso- is central to NAFLD pathogenesis.(14) Our findings genic diet feeding. demonstrated that NOX2 modulated the production of In conclusion, our novel findings suggest that IL-17A, IFN-c, and TNF-a, which are all well-known NOX2 modulates the inflammatory milieu in obesity promoters of NAFLD pathogenesis,(1,14,17,48,49) by (Fig. 7) and, as such, regulates NAFLD pathogenesis. liver-infiltrating immune cells, thus denoting a pivotal Our findings are in agreement with the role of NOX2 role of the NOX2–ROS axis in the pathophysiology of in the modulation of alcoholic fatty liver disease.(50) NAFLD. We also demonstrated that NOX2- Importantly, such insights suggest that pharmacologic dependent effects were not unique to liver-infiltrating T targeting of NOX2 activation and/or function in cells as polarization of naive splenic CD41 T cells to immune cells may represent a viable approach to effector T-cell subtype and cytokine production follow- reducing morbidity of obesity-associated sequelae. ing polyclonal T-cell receptor activation were similarly affected. These data imply that intrinsic NOX2 effects REFERENCES in T cells in combination with the well-established extrinsic signals (e.g., bacterial colonization, lipid parti- 1) Giles DA, Moreno-Fernandez ME, Divanovic S. IL-17 axis tioning) and inflammatory hepatic microenvironment driven inflammation in non-alcoholic fatty liver disease progression. Curr Drug Targets 2015;16:1315-1323. may play vital roles in the regulation of T-cell polariza- 2) Mehal WZ. The Gordian Knot of dysbiosis, obesity and tion and responsiveness in NAFLD. Thus, a definition NAFLD. Nat Rev Gastroenterol Hepatol 2013;10:637-644. of cell-intrinsic and cell-extrinsic mechanisms underly- 3) Polimeni L, Del Ben M, Baratta F, Perri L, Albanese F, Pastori ing NOX2-driven modulation of T-cell polarization D, et al. Oxidative stress: new insights on the association of and inflammatory vigor in NAFLD is clearly warranted.

558 HEPATOLOGY COMMUNICATIONS, Vol. 2, No. 5, 2018 MUKHERJEE ET AL.

non-alcoholic fatty liver disease and atherosclerosis. World J 20) Tilg H, Moschen AR, Roden M. NAFLD and diabetes mellitus. Hepatol 2015;7:1325-1336. Nat Rev Gastroenterol Hepatol 2017;14:32-42. 4) Ganz M, Szabo G. Immune and inflammatory pathways in 21) Cani PD, Amar J, Iglesias MA, Poggi M, Knauf C, Bastelica NASH. Hepatol Int 2013;7(Suppl. 2):771-781. D, et al. Metabolic endotoxemia initiates obesity and insulin 5) Sumida Y, Niki E, Naito Y, Yoshikawa T. Involvement of free resistance. Diabetes 2007;56:1761-1772. radicals and oxidative stress in NAFLD/NASH. Free Radic Res 22) Mas E, Danjoux M, Garcia V, Carpentier S, Segui B, Levade T. 2013;47:869-880. IL-6 deficiency attenuates murine diet-induced non-alcoholic 6) Ray PD, Huang BW, Tsuji Y. Reactive oxygen species (ROS) steatohepatitis. PLoS One 2009;4:e7929. homeostasis and redox regulation in cellular signaling. Cell Signal 23) Costford SR, Castro-Alves J, Chan KL, Bailey LJ, Woo M, 2012;24:981-990. Belsham DD, et al. Mice lacking NOX2 are hyperphagic and 7) Holmstrom KM, Finkel T. Cellular mechanisms and physiologi- store fat preferentially in the liver. Am J Physiol Endocrinol cal consequences of redox-dependent signalling. Nat Rev Mol Metab 2014;306:E1341-E1353. Cell Biol 2014;15:411-421. 24) Strissel KJ, Stancheva Z, Miyoshi H, Perfield JW 2nd, DeFuria 8) Paik YH, Iwaisako K, Seki E, Inokuchi S, Schnabl B, J, Jick Z, et al. Adipocyte death, adipose tissue remodeling, and Osterreicher CH, et al. The nicotinamide adenine dinucleotide obesity complications. Diabetes 2007;56:2910-2918. phosphate oxidase (NOX) homologues NOX1 and NOX2/ 25) Lynch CM, Kinzenbaw DA, Chen X, Zhan S, Mezzetti E, gp91(phox) mediate hepatic fibrosis in mice. Hepatology 2011; Filosa J, et al. Nox2-derived superoxide contributes to cerebral 53:1730-1741. vascular dysfunction in diet-induced obesity. Stroke 2013;44: 9) Sukumar P, Viswambharan H, Imrie H, Cubbon RM, 3195-3201. Yuldasheva N, Gage M, et al. Nox2 NADPH oxidase has a crit- 26) Coats BR, Schoenfelt KQ, Barbosa-Lorenzi VC, Peris E, Cui ical role in insulin resistance-related endothelial cell dysfunction. C, Hoffman A, et al. Metabolically activated adipose tissue mac- Diabetes 2013;62:2130-2134. rophages perform detrimental and beneficial functions during 10) Souto Padron de Figueiredo A, Salmon AB, Bruno F, Jimenez diet-induced obesity. Cell Rep 2017;20:3149-3161. F, Martinez HG, Halade GV, et al. Nox2 mediates skeletal mus- 27) Minokoshi Y, Kahn CR, Kahn BB. Tissue-specific ablation of cle insulin resistance induced by a high fat diet. J Biol Chem the GLUT4 glucose transporter or the insulin receptor challenges 2015;290:13427-13439. assumptions about insulin action and glucose homeostasis. J Biol 11) Pepping JK, Freeman LR, Gupta S, Keller JN, Bruce-Keller AJ. Chem 2003;278:33609-33612. NOX2 deficiency attenuates markers of adiposopathy and brain 28) Softic S, Gupta MK, Wang GX, Fujisaka S, O’Neill BT, Rao injury induced by high-fat diet. Am J Physiol Endocrinol Metab TN, et al. Divergent effects of glucose and fructose on hepatic 2013;304:E392-E404. lipogenesis and insulin signaling. J Clin Invest 2017;127:4059- 12) Divanovic S, Dalli J, Jorge-Nebert LF, Flick LM, Galvez- 4074. Peralta M, Boespflug ND, et al. Contributions of the three 29) de Souza AH, Santos LRB, Roma LP, Bensellam M, Carpinelli CYP1 monooxygenases to pro-inflammatory and inflammation- AR, Jonas JC. NADPH oxidase-2 does not contribute to beta- resolution lipid mediator pathways. J Immunol 2013;191:3347- cell glucotoxicity in cultured pancreatic islets from C57BL/6J 3357. mice. Mol Cell Endocrinol 2017;439:354-362. 13) Divanovic S, Trompette A, Atabani SF, Madan R, Golenbock 30) Mofrad P, Contos MJ, Haque M, Sargeant C, Fisher RA, DT, Visintin A, et al. Negative regulation of Toll-like receptor 4 Luketic VA, et al. Clinical and histologic spectrum of nonalco- signaling by the Toll-like receptor homolog RP105. Nat Immu- holic fatty liver disease associated with normal ALT values. Hep- nol 2005;6:571-578. atology 2003;37:1286-1292. 14) Giles DA, Moreno-Fernandez ME, Stankiewicz TE, 31) Nalbantoglu IL, Brunt EM. Role of liver biopsy in nonalcoholic Graspeuntner S, Cappelletti M, Wu D, et al. Thermoneutral fatty liver disease. World J Gastroenterol 2014;20:9026-9037. housing exacerbates nonalcoholic fatty liver disease in mice and 32) Nativ NI, Yarmush G, So A, Barminko J, Maguire TJ, Schloss allows for sex-independent disease modeling. Nat Med 2017;23: R, et al. Elevated sensitivity of macrosteatotic hepatocytes to 829-838. Erratum in: Nat Med 2017;23:1241. hypoxia/reoxygenation stress is reversed by a novel defatting pro- 15) Cappelletti M, Presicce P, Lawson MJ, Chaturvedi V, tocol. Liver Transpl 2014;20:1000-1011. Stankiewicz TE, Vanoni S, et al. Type I interferons regulate sus- 33) Furukawa S, Fujita T, Shimabukuro M, Iwaki M, Yamada Y, ceptibility to inflammation-induced preterm birth. JCI Insight Nakajima Y, et al. Increased oxidative stress in obesity and its 2017;2:e91288. impact on metabolic syndrome. J Clin Invest 2004;114:1752- 16) Weisberg SP, McCann D, Desai M, Rosenbaum M, Leibel RL, 1761. Ferrante AW Jr. Obesity is associated with macrophage accumu- 34) Paquissi FC. Immune imbalances in non-alcoholic fatty liver dis- lation in adipose tissue. J Clin Invest 2003;112:1796-1808. ease: from general biomarkers and neutrophils to interleukin-17 17) Harley IT, Stankiewicz TE, Giles DA, Softic S, Flick LM, axis activation and new therapeutic targets. Front Immunol 2016; Cappelletti M, et al. IL-17 signaling accelerates the progression 7:490. of nonalcoholic fatty liver disease in mice. Hepatology 2014;59: 35) Gadd VL, Skoien R, Powell EE, Fagan KJ, Winterford C, 1830-1839. Horsfall L, et al. The portal inflammatory infiltrate and ductular 18) Giles DA, Moreno-Fernandez ME, Stankiewicz TE, Cappelletti reaction in human nonalcoholic fatty liver disease. Hepatology M, Huppert SS, Iwakura Y, et al. Regulation of inflammation by 2014;59:1393-1405. IL-17A and IL-17F modulates non-alcoholic fatty liver disease 36) Pepping JK, Vandanmagsar B, Fernandez-Kim SO, Zhang J, pathogenesis. PLoS One 2016;11:e0149783. Mynatt RL, Bruce-Keller AJ. Myeloid-specific deletion of 19) Giles DA, Ramkhelawon B, Donelan EM, Stankiewicz TE, NOX2 prevents the metabolic and neurologic consequences of Hutchison SB, Mukherjee R, et al. Modulation of ambient tem- high fat diet. PLoS One 2017;12:e0181500. perature promotes inflammation and initiates atherosclerosis in 37) Garcia-Ruiz I, Solis-Munoz P, Fernandez-Moreira D, Grau M, wild type C57BL/6 mice. Mol Metab 2016;5:1121-1130. Munoz-Yague T, Solis-Herruzo JA. NADPH oxidase is

559 MUKHERJEE ET AL. HEPATOLOGY COMMUNICATIONS, May 2018

implicated in the pathogenesis of oxidative phosphorylation dys- 45) Petersen MC, Vatner DF, Shulman GI. Regulation of hepatic function in mice fed a high-fat diet. Sci Rep 2016;6:23664. glucose metabolism in health and disease. Nat Rev Endocrinol 38) Ye D, Li FY, Lam KS, Li H, Jia W, Wang Y, et al. Toll-like 2017;13:572-587. receptor-4 mediates obesity-induced non-alcoholic steatohepatitis 46) Loh K, Deng H, Fukushima A, Cai X, Boivin B, Galic S, et al. through activation of X-box binding protein-1 in mice. Gut Reactive oxygen species enhance insulin sensitivity. Cell Metab 2012;61:1058-1067. 2009;10:260-272. 39) Miura K, Yang L, van Rooijen N, Brenner DA, Ohnishi H, 47) Del Ben M, Polimeni L, Carnevale R, Bartimoccia S, Nocella C, Seki E. Toll-like receptor 2 and palmitic acid cooperatively Baratta F, et al. NOX2-generated oxidative stress is associated contribute to the development of nonalcoholic steatohepatitis with severity of ultrasound liver steatosis in patients with non- through inflammasome activation in mice. Hepatology 2013; alcoholic fatty liver disease. BMC Gastroenterol 2014;14:81. 57:577-589. 48) Alisi A, Carpino G, Oliveira FL, Panera N, Nobili V, Gaudio 40) Yang AM, Inamine T, Hochrath K, Chen P, Wang L, Llorente E. The role of tissue macrophage-mediated inflammation on C, et al. Intestinal fungi contribute to development of alcoholic NAFLD pathogenesis and its clinical implications. Mediators liver disease. J Clin Invest 2017;127:2829-2841. Inflamm 2017;2017:8162421. 41) Leung C, Rivera L, Furness JB, Angus PW. The role of the gut 49) Narayanan S, Surette FA, Hahn YS. The immune landscape in microbiota in NAFLD. Nat Rev Gastroenterol Hepatol 2016;13: nonalcoholic steatohepatitis. Immune Netw 2016;16:147-158. 412-425. 50) Wang M, Frasch SC, Li G, Feng D, Gao B, Xu L, et al. Role 42) Zhu L, Baker SS, Gill C, Liu W, Alkhouri R, Baker RD, et al. of gp91phox in hepatic macrophage programming and alcoholic Characterization of gut microbiomes in nonalcoholic steatohepa- liver disease. Hepatol Commun 2017;1:765-779. titis (NASH) patients: a connection between endogenous alcohol and NASH. Hepatology 2013;57:601-609. Author names in bold designate shared co-first 43) Kozak LP, Koza RA, Anunciado-Koza R. Brown fat thermogenesis and body weight regulation in mice: relevance to humans. authorship. Int J Obes (Lond) 2010;34(Suppl. 1):S23-S27. 44) Chouchani ET, Kazak L, Jedrychowski MP, Lu GZ, Erickson BK, Szpyt J, et al. Mitochondrial ROS regulate thermogenic Supporting Information energy expenditure and sulfenylation of UCP1. Nature 2016;532: 112-116. Erratum in: Nature 2016;536:360. Additional Supporting Information may be found at onlinelibrary.wiley.com/doi/10.1002/hep4.1162/full.

560 ARTICLE DOI: 10.1038/s41467-018-05870-6 OPEN Hepatic Ago2-mediated RNA silencing controls energy metabolism linked to AMPK activation and obesity-associated pathophysiology

Cai Zhang1,2, Joonbae Seo 1, Kazutoshi Murakami1, Esam S.B. Salem1,3, Elise Bernhard1, Vishnupriya J. Borra1, Kwangmin Choi4, Celvie L. Yuan1, Calvin C. Chan5, Xiaoting Chen6, Taosheng Huang7,8, Matthew T. Weirauch 6,7,9,10, Senad Divanovic7,11, Nathan R. Qi12, Hala Einakat Thomas13, Carol A. Mercer 13, Haruhiko Siomi14 & Takahisa Nakamura 1,7,9 1234567890():,;

RNA silencing inhibits mRNA translation. While mRNA translation accounts for the majority of cellular energy expenditure, it is unclear if RNA silencing regulates energy homeostasis. Here, we report that hepatic Argonaute 2 (Ago2)-mediated RNA silencing regulates both intrinsic energy production and consumption and disturbs energy metabolism in the pathogenesis of obesity. Ago2 regulates expression of specific miRNAs including miR-802, miR- 103/107, and miR-148a/152, causing metabolic disruption, while simultaneously suppressing the expression of genes regulating glucose and lipid metabolism, including Hnf1β, Cav1, and Ampka1. Liver-specific Ago2-deletion enhances mitochondrial oxidation and ATP consumption associated with mRNA translation, which results in AMPK activation, and improves obesity-associated pathophysiology. Notably, hepatic Ago2-deficiency improves glucose metabolism in conditions of insulin receptor antagonist treatment, high-fat diet challenge, and hepatic AMPKα1-deletion. The regulation of energy metabolism by Ago2 provides a novel paradigm in which RNA silencing plays an integral role in determining basal metabolic activity in obesity-associated sequelae.

1 Division of Endocrinology, Cincinnati Children’s Hospital Medical Center, Cincinnati, OH, USA. 2 Department of Pediatrics, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, China. 3 Department of Pharmacology and Systems Physiology, University of Cincinnati College of Medicine, Cincinnati, OH, USA. 4 Division of Experimental Hematology and Cancer Biology, Cincinnati Children’s Hospital Medical Center, Cincinnati, OH, USA. 5 Medical Scientist Training Program, Immunology Graduate Program, Cincinnati Children’s Hospital Medical Center and University of Cincinnati College of Medicine, Cincinnati, OH, USA. 6 Division of Biomedical Informatics, Cincinnati Children’s Hospital Medical Center, Cincinnati, OH, USA. 7 Department of Pediatrics, University of Cincinnati College of Medicine, Cincinnati, OH, USA. 8 Division of Human Genetics, Cincinnati Children’s Hospital Medical Center, Cincinnati, OH, USA. 9 Division of Developmental Biology, Cincinnati Children’s Hospital Medical Center, Cincinnati, OH, USA. 10 Center for Autoimmune Genomics and Etiology, Cincinnati Children’s Hospital Medical Center, Cincinnati, OH, USA. 11 Division of Immunobiology, Cincinnati Children’s Hospital Medical Center, Cincinnati, OH, USA. 12 Department of Internal Medicine, University of Michigan Medical School, Ann Arbor, MI, USA. 13 Division of Hematology-Oncology, Department of Internal Medicine, University of Cincinnati College of Medicine, Cincinnati, OH, USA. 14 Department of Molecular Biology, Keio University School of Medicine, Tokyo, Japan. These authors contributed equally: Cai Zhang, Joonbae Seo, Kazutoshi Murakami. Correspondence and requests for materials should be addressed to T.N. (email: [email protected])

NATURE COMMUNICATIONS | (2018)9:3658 | DOI: 10.1038/s41467-018-05870-6 | www.nature.com/naturecommunications 1 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-05870-6

he worldwide prevalence of obesity has reached pandemic feature in obesity and the pathogenesis of obesity-associated proportions, bringing with it a host of associated diseases, sequelae. Tsuch as type 2 diabetes (T2D) and non-alcoholic steatohepatitis (NASH)1,2. Obesity develops when energy intake chronically exceeds total energy expenditure. Basal metabolic rate Results represents the largest component of total energy expenditure, of Hepatic Ago2 regulates expression of specific miRNAs which the liver is a major organ for energy consumption3. Protein involved in energy metabolism. Ago1 and Ago2 are the pre- biosynthesis is one of the most energy-consuming cellular dominant Ago family members expressed in the liver25. To processes in the liver, accounting for ~20–30% of total energy investigate if RNA silencing is associated with energy and meta- consumption4,5. However, despite abundant supply of energy bolic homeostasis, we generated liver-specific Ago1-deficient sources and a robust activation of the mammalian target of (L-Ago1 KO: Ago1fl/fl Alb-CreTg/0) and Ago2-deficient (L-Ago2 rapamycin complex (mTORC) pathway, a main driver of protein KO: Ago2fl/fl Alb-CreTg/0) mice (Supplementary Fig. 1a and b). On synthesis6–9, the liver-driven energy consumption robustly control diet (CD), both L-Ago1 KO and L-Ago2 KO mice gained declined in obesity due to, at least in part, insufficient protein weight similarly to their controls, L-Ago1 WT (Ago1fl/fl Alb-Cre0/0) biosynthesis. Suppression of hepatic protein synthesis leads to or L-Ago2 WT (Ago2fl/fl Alb-Cre0/0), and showed no obvious further accumulation of energy sources associated with obesity- abnormalities during development or in adulthood. In addition, associated pathophysiology, however, the exact mechanism(s) of levels of serum alanine aminotransferase (ALT) were comparable insufficient protein biosynthesis remains unclear. Hence, defining between the groups (Supplementary Fig. 1c and d). These results such molecular mechanism(s) could provide a novel therapeutic suggest that general functions of the liver are likely unaffected by approach that alters energy balance in obesity and modulates the the absence of Ago1 or Ago2 during development in regular pathogenesis of associated sequelae. feeding conditions. Recent studies have revealed significant roles for microRNA Among the Ago proteins, Ago2 uniquely possesses a slicer (miRNA)-mediated events in the development and progression of activity known to contribute to the expression of specific obesity and its associated sequelae10,11. Global dysregulation of miRNA21–23 and mRNA cleavage18,26. In the Ago2-deficient miRNA expression is triggered in the human and murine obese livers, expression levels of Ago1 are increased (Supplementary liver, leading to the induction of the vast majority of miRNAs, Fig. 1b), and therefore Ago1 may compensate for Ago2’s non- including miR-802, miR-103/miR-107, and miR-148a that dete- slicer activity-dependent function. To gain insight into the riorate glucose and lipid metabolism in obesity12–16. As miRNA specific roles of Ago2’s activity in the regulation of liver function, generally inhibits the translation of target mRNAs through RNA we first assessed the effect of hepatic Ago2-deficiency on miRNA silencing, it is reasonable to hypothesize that these induced expression profile. Expression profile analyses revealed that the miRNAs may contribute to suppression of protein biosynthesis expression levels of 25 miRNAs were significantly reduced in L- and its associated energy expenditure in obese liver. However, Ago2 KO liver, while 8 miRNAs were significantly increased there remain fundamental questions concerning why and how (Fig. 1a, b). Among these significant miRNAs, miR-148a is one of these miRNAs are concurrently induced in the obese condition the most abundant miRNAs expressing in the WT liver (Fig. 1b and whether RNA silencing is integrated into an elaborate and Supplementary Fig. 2a). In addition, 17 out of 33 total adaptive program that cells can elicit to balance anabolic and significant miRNAs abundantly expressed were in the top 10 catabolic processes dependent on energy and metabolic statuses. percent of miRNAs expressing in the WT liver (Fig. 1c and If RNA silencing plays a role in protein biosynthesis-associated Supplementary Table 1). Intriguingly, this group contained energy metabolism, one would anticipate that individual com- miRNAs known to be associated with metabolic diseases (MD- ponent(s) of miRNA-regulatory machinery in the liver may miRNAs) and detrimental to glucose and lipid metabolism, impinge on metabolic regulation, and that a nutrient challenge including miR-802, miR-103/107, miR-130a, and miR- might accentuate the consequences of this regulation. 148a12,13,27–29, which were downregulated in the L-Ago2 KO Argonaute (Ago) family proteins are the main components of liver (Fig. 1c). We additionally utilized the list of significant the RNA-induced silencing complex (RISC) that carries out RNA miRNAs altered in Ago2 KO liver through the miRNA silencing. Upon loading of Ago proteins with mature miRNAs enrichment pathway analysis by the biological processes and produced by the endoribonuclease Dicer17,18, RISC represses the molecular function categories (Fig. 1d). Hepatic Ago2 deficiency expression of targeted mRNA through RNA silencing18–20. affected the functional clades associated with energy metabolism Amongst all Ago proteins, Ago2 specifically possesses an endor- including fatty acid biosynthesis, AMPK signaling, protein ibonuclease (“slicer”) activity that generates a specific mature processing, and insulin signaling. Taken together, these results miRNA and cleaves targeted mRNAs in mammals18,21–24. To imply the unique role of Ago2 in the expressional regulation of a study the role of RNA silencing in hepatic energy homeostasis, we specific repertoire of MD-miRNAs for energy metabolism. comprehensively evaluated the role of hepatic Ago1 and Ago2, as Consistent with these observations, the reduction of MD- analyses of these core RISC components might lead to funda- miRNAs in L-Ago2 KO liver is accompanied by increased mental insights into the link of RNA silencing with energy expression of their known target mRNAs, such as hepatocyte metabolism. This study demonstrates that hepatic Ago2-mediated nuclear factor 1 homeobox B (Hnf1β), caveolin-1 (Cav1), RNA silencing regulates energy expenditure during the course of peroxisome proliferator-activated receptor gamma coactivator obesity and its inactivation protects from obesity-associated glu- 1α (Pgc1α), and low-density lipoprotein receptor (Ldlr), which cose intolerance and hepatic steatosis in mice. Importantly, we regulates energy metabolism (Fig. 1e)12,13,29. discover novel roles of Ago2 in orchestrating the expression of a To examine the mechanism by which hepatic Ago2 regulates subset of miRNAs, including miR-802, miR-103/107, and miR- expression of specific miRNAs, we measured the expression levels 148a, and in the regulation of AMP-activated protein kinase of each mature and primary miRNA (pri-miRNA) employing (AMPK) activation linked to protein biosynthesis-mediated TaqMan probe-based gene expression analysis. Mature miRNA energy consumption. This Ago2-mediated RNA silencing levels of miR-802, miR-107/miR-103, miR-130a, and miR-148a/ is a critical mechanism that connects the dots between 148b/152, were reduced in L-Ago2 KO liver (Fig. 1f), despite protein translation, energy production and consumption, and intact expression levels of their pri-miRNAs (Fig. 1g). These AMPK activity—disruption of such events is a well-recognized results suggest that the miRNA maturation process is impaired in

2 NATURE COMMUNICATIONS | (2018)9:3658 | DOI: 10.1038/s41467-018-05870-6 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-05870-6 ARTICLE

abL-Ago2 L-Ago2 c WT KO mmu-miR-148a-3p 18 mmu-miR-148b-3p 4.5 mmu-miR-101a-3p L-Ago2 WT mmu-miR-802-3p * L-Ago2 KO mmu-miR-802-5p 16 ** ** 4 mmu-miR-101b-3p mmu-miR-27a-3p ** 3.5 mmu-miR-27b-3p ** mmu-miR-425-5p 14 ** ** mmu-miR-140-3p 3 mmu-miR-93-5p * mmu-miR-130a-3p * mmu-miR-29b-3p 12 * 2.5 * * * mmu-miR-185-5p * * mmu-miR-1839-5p 2 mmu-miR-194-5p 10 * ** mmu-miR-365-3p * ** ** mmu-miR-103-3p ** ** 1.5 mmu-miR-107-3p ** ** ** ** ** mmu-miR-17-5p 8 ** ** mmu-miR-1981-3p ** 1 mmu-miR-320-3p ** changes (WT/KO) Fold mmu-miR-101a-5p ** ** mmu-miR-141-3p 6 0.5 mmu-miR-200b-3p

mmu-miR-181a-5p Log2 normalized counts (DESeq2) mmu-miR-30e-5p 0 mmu-miR-125b-5p 4 mmu-miR-150-5p mmu-miR-411-5p mmu-miR-122-3p mmu-miR-22-5p Z -score mmu-miR-93-5p mmu-miR-802-5p mmu-miR-107-3p mmu-miR-194-5p mmu-miR-103-3p mmu-miR-27b-3p mmu-miR-27a-3p mmu-miR-140-3p mmu-miR-122-3p mmu-miR-30e-5p mmu-miR-93-5p mmu-miR-17-5p mmu-miR-22-5p mmu-miR-130a-3p mmu-miR-101b-3p mmu-miR-101a-3p mmu-miR-148a-3p mmu-miR-148b-3p mmu-miR-125b-5p mmu-miR-181a-5p mmu-miR-27b-3p mmu-miR-27a-3p mmu-miR-194-5p mmu-miR-107-3p mmu-miR-802-5p mmu-miR-103-3p mmu-miR-140-3p mmu-miR-425-5p mmu-miR-29b-3p mmu-miR-365-3p mmu-miR-802-3p mmu-miR-141-3p mmu-miR-320-3p mmu-miR-185-5p mmu-miR-30e-5p mmu-miR-122-3p mmu-miR-150-5p mmu-miR-411-5p –1.5 0 1.5 mmu-miR-148a-3p mmu-miR-101b-3p mmu-miR-101a-3p mmu-miR-130a-3p mmu-miR-148b-3p mmu-miR-1839-5p mmu-miR-101a-5p mmu-miR-148b-5p mmu-miR-1981-3p mmu-miR-125b-5p mmu-miR-181a-5p mmu-miR-200b-3p

d Down-miRNAs Up-miRNAs Fatty acid biosynthesis e f g Endocytosis L-Ago2 WT L-Ago2 WT FoxO signaling pathway L-Ago2 WT MAPK signaling pathway 1.5 L-Ago2 KO L-Ago2 KO Hippo signaling pathway 2.5 L-Ago2 KO 1.6 * PI3K-Akt signaling pathway TGF-beta signaling pathway ** * 1.4 * Ras signaling pathway 2 Wnt signaling pathway 1.2 Hedgehog signaling pathway ** ** 1 mRNA surveillance pathway 1 ErbB signaling pathway 1.5 AMPK signaling pathway ** 0.8 cAMP signaling pathway ** mTOR signaling pathway 1 ** ** 0.6 Glycosaminoglycan biosynthesis 0.5 ** ** Insulin signaling pathway ** Lysine degradation 0.4 Fatty acid metabolism 0.5 cGMP-PKG signaling pathway mRNA levels Target 0.2 Phosphatidylinositol signaling system miRNA levels Relative Inositol phosphate metabolism 0 0 pri-miRNA levels Relative 0 Ubiquitin mediated proteolysis N-Glycan biosynthesis β α Ldlr Protein processing in ER Cav1 Glycosphingolipid biosynthesis Hnf1 Pgc1 Sphingolipid signaling pathway miR-802miR-107miR-103 miR-152 Glycosphingolipid biosynthesis miR-130amiR-148amiR-148b Type II diabetes mellitus pri-miR-802pri-miR-107 pri-miR-130apri-miR-148apri-miR-148bpri-miR-152 TNF signaling pathway pri-miR-103-1pri-miR-103-2 8 6 4 2 0 2 4 6 8 –Log10 (p -value)

Fig. 1 Effects of hepatic Ago2-deficiency on MD-miRNA expression in the liver. a–c A heatmap diagram illustrating the differential expression of hepatic mature miRNAs (a), raw counts of significant miRNAs normalized by DESeq2 and transformed by Log2 (b), and fold changes of significant miRNAs whose expression levels are in the top 10 percent in the WT liver (c) in the liver of L-Ago2 WT (n = 3) and L-Ago2 KO (n = 3) mice fed NCD at 9 weeks of age. Significant miRNAs differentially expressed between L-Ago2 WT and L-Ago2 KO groups were identified using DESeq2 (|fold change| > 2x and adjusted p < 0.05) and plotted as a heatmap using z-score. d Metabolic pathway enrichment analysis of miRNAs significantly downregulated (blue) and upregulated (red) miRNAs in the liver of L-Ago2 KO mice fed NCD at 9 weeks of age. These miRNAs were queried to calculate the most enriched KEGG pathways using DIANA-mirPath web-server (p < 0.05 and MicroT threshold < 0.8). Pathways unrelated to hepatic functions were excluded in this pyramid plot. e–g Expression levels of MD-miRNAs’ target mRNAs (e), selective MD-miRNAs (f), and their pri-miRNAs (g) in the liver of L-Ago2 WT (n = 8) and L-Ago2 KO (n = 8) mice fed NCD at 25 weeks of age. Data are shown as the mean ± SEM. *p < 0.05, **p < 0.01 the Ago2-deficient liver. To further confirm this regulation, we guide and passenger strands32. The miRNAs with reduced utilized Ago2-deficient mouse embryonic fibroblasts (MEFs) expression in L-Ago2 KO liver tended to have shorter loop sizes reconstituted with wild type Ago2 (Ago2 WT) or slicer- than those induced by L-Ago2 KO liver, and had no mis-matching defective mutant Ago2 (Ago2 D669A, or “DA,” containing an at positions 10 or 11 (Supplementary Fig. 2d–f). This information aspartate to alanine substitution at residue 669)22–24. Expression suggests that there may be structural similarities among levels of these MD-miRNAs were increased by reconstitution of MD-miRNAs that require Ago2 for maturation. Ago2 WT (Supplementary Fig. 2b). Importantly, Ago2 D669A We next asked if any of the 25 significant miRNAs, Ago1, or mutant only partially induced expression of key MD-miRNAs Ago2 might have disease associations proximal to their including miR-107, miR-103, and miR-130a, while expression of orthologous human miRNAs and genes. To this end, we first miR-148b required Ago2 but not its slicer activity (Supplemen- identified mouse-human orthologous miRNAs using data from tary Fig. 2b). While the loss of Dicer also caused a reduction of miRBase33 and miROrtho34, identifying a total of 27 human MD-miRNAs (Supplementary Fig. 2c), these results support a miRNAs (some mouse miRNAs do not have clear orthologs or crucial role of Ago2 in the expression of a subset of MD-miRNAs. map to two human miRNAs) (Supplementary Table 2). We then While Dicer recognizes the 5′ phosphate end and 2-nucleotide 3′ examined a large collection of genetic variants associated with 213 overhang structure of precursor miRNA for precise and effective diseases and phenotypes collected from the National Human biogenesis of miRNAs30,31, recent studies have provided different Genome Research Institute (NHGRI) Catalog of Published mechanistic insights into the Ago2-mediated processing of Genome-Wide Association Studies (GWAS)35 that have been miRNA. One of the proposed characteristics of miRNAs processed expanded to include additional variants in strong linkage by Ago2 is that their precursors have a relatively shorter loop size disequilibrium (r2 > 0.8) with the tagged variants36. This analysis that likely prevents recognition by Dicer. Moreover, these revealed that several of the orthologous human miRNAs have precursors have no mis-matching at position 10 or 11 between proximal GWAS signal, often for relevant phenotypes

NATURE COMMUNICATIONS | (2018)9:3658 | DOI: 10.1038/s41467-018-05870-6 | www.nature.com/naturecommunications 3 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-05870-6

ab c 500 200 250 ) ) ) –1

–1 200 400 –1 150 ** 300 150 ** 100 200 100 ** 50 100 50 Blood glucose (mg dl

Blood glucose (mg dL L-Ago2 WT L-Ago2 WT Blood glucose (mg dL L-Ago2 WT L-Ago2 KO L-Ago2 KO L-Ago2 KO 0 0 0 0 30 60 90120 0 30 60 90 120 0 30 60 90 120 Time after IP glucose (min) Time after IP insulin (min) Time after IP pyruvate (min)

defgGlucose production Glycolytic capacity Pyruvate oxidation ATP/ADP 2 ** 3 ** ** ** 7 ** 40 ** ** 1.8 ** ** ** ** 35 ** protein) 1.6 6 protein) 2.5 –1 –1 30 g 1.4 g µ 5 µ

2 ** 25 –1

1.2 –1 4 1 20 1.5 ** 0.8 3 15 1

Relative levels Relative 0.6 2 10 levels Relative 0.4 0.5 0.2 1 5 OCR (pMole min 0 ECAR (mPH min 0 0 0 – 100 200 –––– 100 200 –––Bt-cAMP Glucose –+–+ Pyruvate–+–+ Pyruvate –+–+ ––––100 200 ––––100 200 pCPT-cAMP L-Ago2 L-Ago2 L-Ago2 L-Ago2 L-Ago2 L-Ago2 L-Ago2 WT L-Ago2 KO WT KO WT KO WT KO

Fig. 2 Hepatic Ago2-deﬁciency improves glucose metabolism. a Glucose tolerance test performed in L-Ago2 WT (n = 10) and KO (n = 8) mice fed NCD at 20 weeks of age. b Insulin tolerance test performed in L-Ago2 WT (n = 9) and KO (n = 6) mice fed NCD at 21 weeks of age. c Pyruvate tolerance test performed in L-Ago2 WT (n = 9) and KO (n = 10) mice fed NCD at 24 weeks of age. d Glucose production in primary hepatocytes isolated from L-Ago2 WT and L-Ago2 KO incubated in the absence (n = 4 for L-Ago2 WT and L-Ago2 KO, respectively) or presence of 100 or 200 μM Bt-cAMP or pCPT-cAMP (n = 2 for L-Ago2 WT and L-Ago2 KO, respectively). e Extracellular acidiﬁcation (ECAR) in the absence or presence of 10 mM glucose in primary hepatocytes isolated from L-Ago2 WT (n = 3 for control, n = 6 for glucose) and L-Ago2 KO mice (n = 3 for control, n = 6 for glucose). f Mitochondrial oxygen consumption rate (OCR) in the absence or presence of 2 mM pyruvate in primary hepatocytes isolated from L-Ago2 WT (n = 3 for control and pyruvate, respectively) and L-Ago2 KO mice (n = 3 for control and pyruvate, respectively). g The ATP/ADP ratio in primary hepatocytes isolated from L- Ago2 WT and L-Ago2 KO mice in the absence (n = 6 for L-Ago2 WT and L-Ago2 KO, respectively) or presence of 5 mM pyruvate (n = 6 for L-Ago2 WT and L-Ago2 KO, respectively). Data are shown as the mean ± SEM. *p < 0.05, **p < 0.01

(Supplementary Table 3). For example, a genetic variant catabolic capacities of hepatocytes. To examine the glycolytic rate, (rs6953596) located 245 bases away from the gene encoding the extracellular acidification rate (ECAR) was determined in miR-148a is strongly associated with body mass index (BMI) in primary hepatocytes upon addition of glucose. In the presence of African Americans and thus might act by altering gene regulatory oligomycin, Ago2-deficient hepatocytes showed a higher increase mechanisms controlling the expression of miR-148a. In addition, in ECAR compared to controls (Fig. 2e). To determine whether while there is no detectable GWAS signal near Ago1, there is Ago2 regulates oxidation of pyruvate, we measured mitochondrial suggestive GWAS signal (p = 10−6) located within an intron of oxygen consumption rate (OCR) in WT and Ago2-deficient Ago2 for “Thiazide-induced adverse metabolic effects in hyper- hepatocytes in the presence of pyruvate. Upon the addition of the tensive patients” in African Americans37. These results suggest a protonophore carbonyl cyanide 4-(trifluoromethoxy) phenylhy- possible association of Ago2 and miRNAs whose expression is drazone (FCCP), Ago2-deficient hepatocytes greatly upregulated regulated by Ago2 with human metabolic diseases. oxygen consumption compared to WT controls (Fig. 2f). In a complementary approach, we also measured ATP content/ADP content in these hepatocytes in the absence or presence of pyr- Inactivation of hepatic Ago2 improves systemic glucose fi metabolism. Considering the miRNA enrichment pathway ana- uvate. The basal ATP/ADP ratio in Ago2-de cient hepatocytes lysis indicated that hepatic Ago2 is implicated in glucose meta- was higher than that in WT controls, and this difference was even bolism, we then investigated Ago2’s role in regulating this regard. greater upon addition of pyruvate (Fig. 2g). Taken together, these We observed that L-Ago2 KO mice fed normal chow diet (NCD) data indicate that hepatic Ago2 functions to enhance gluconeo- exhibited enhanced capacities for glucose metabolism, as assessed genesis and suppress glucose oxidation while its inactivation by glucose, insulin, and pyruvate tolerance tests (GTT, ITT, and results in increased glucose-driven energy production. PTT) after 20 weeks of age (Fig. 2a–c). These results suggest that hepatic Ago2 deficiency improves insulin sensitivity and inhibits Hepatic Ago2 regulates glucose metabolism in insulin insuffi- gluconeogenesis, leading to glucose tolerance. To investigate how ciency. If hepatic Ago2-deficiency improves systemic glucose Ago2 regulates glucose metabolism, we first examined capacities metabolism by suppressing gluconeogenesis and accelerating of hepatic gluconeogenesis in Ago2-deficient primary hepato- glucose oxidation in the liver, other diabetic conditions may also cytes. Glucose production was similarly induced between geno- be improved by hepatic Ago2-deficiency. To examine this pos- types (Fig. 2d). We then asked if Ago2 regulates the fundamental sibility, we employed a pharmacological model by administering

4 NATURE COMMUNICATIONS | (2018)9:3658 | DOI: 10.1038/s41467-018-05870-6 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-05870-6 ARTICLE

a 500 AUCbc Insulin Glycogen 1.8 ** 30 **1.2 ** )

–1 400 1.6 25 1 1.4 * )

300 1.2 –1 20 0.8 1 15 0.6 200 0.8 L-Ago2 WT_PBS 0.6 10 0.4 Relative levels Relative Insulin (ng ml 100 L-Ago2 WT_S961 0.4 contents Relative Blood glucose (mg dL L-Ago2 KO_PBS 5 0.2 L-Ago2 KO_S961 0.2 0 0 0 0 0 15 30 60 90 120 L-Ago2 WT PBS L-Ago2 KO PBS Time after IP glucose (min) L-Ago2 WT S961 L-Ago2 KO S961

d L-Ago2 WT L-Ago2 KO e PBS S961 PBS S961 2 * * * * mmu-miR-1948-3p mmu-miR-222-3p * * * * mmu-miR-320-3p mmu-miR-421-3p * * * * mmu-miR-186-5p mmu-miR-365-3p mmu-miR-1981-5p 1.5 mmu-miR-103-3p mmu-miR-107-3p mmu-miR-1839-3p mmu-miR-574-5p mmu-miR-6964-3p mmu-miR-32-5p L-Ago2 WT PBS mmu-miR-93-5p (I) mmu-miR-292b-5p 1 mmu-miR-339-5p L-Ago2 WT S961 mmu-miR-330-5p mmu-miR-18a-5p mmu-miR-802-3p mmu-miR-802-5p L-Ago2 KO PBS mmu-miR-193a-3p mmu-miR-7219-3p 0.5 mmu-miR-107-5p miRNA levels Relative L-Ago2 KO S961 mmu-miR-6978-3p mmu-miR-17-3p mmu-miR-3105-3p mmu-miR-128-3p mmu-miR-130a-3p mmu-miR-484 0 mmu-miR-17-5p mmu-miR-194-5p mmu-miR-425-5p miR-802 miR-107 miR-103 miR-130a mmu-miR-1981-3p mmu-miR-33-5p mmu-miR-106b-3p mmu-miR-182-5p f (II) mmu-miR-1191b-5p mmu-miR-203-3p ** mmu-miR-409-3p 2.5 mmu-miR-381-3p ** mmu-miR-127-3p * mmu-miR-541-5p (III) mmu-miR-300-3p * mmu-miR-337-5p * mmu-miR-376b-3p 2 * * mmu-let-7g-3p * * mmu-let-7a-1-3p mmu-let-7c-2-3p * mmu-miR-26b-3p mmu-miR-145a-5p mmu-miR-22-5p 1.5 mmu-miR-547-3p mmu-miR-92a-1-5p ** mmu-miR-30c-2-3p mmu-miR-30e-5p mmu-let-7f-1-3p mmu-let-7f-2-3p 1 mmu-miR-99a-3p mmu-miR-30c-1-3p mmu-miR-664-5p mmu-miR-30d-3p mmu-miR-351-5p (IV) mmu-miR-194-2-3p 0.5 mmu-miR-181b-5p mRNA levels Relative mmu-miR-192-5p mmu-miR-3535 mmu-miR-181a-5p mmu-miR-322-3p mmu-miR-181a-1-3p 0 mmu-miR-148a-5p mmu-miR-30a-5p mmu-miR-146a-5p Hnf1β Cav1 Pgc1α Pparα Tfam Cs G6Pase Ampka1 Ampka2 mmu-miR-10b-5p mmu-miR-10a-5p mmu-miR-143-3p L-Ago2 WT PBS L-Ago2 WT PBS mmu-let-7b-3p mmu-miR-184-3p mmu-miR-148a-3p L-Ago2 WT S961 L-Ago2 WT S961 mmu-miR-148b-3p mmu-miR-152-3p Z -score –20 2

Fig. 3 Hepatic Ago2-deficiency prevents S961-induced acute glucose intolerance. a Glucose tolerance tests at one-week post treatment of S961 or phosphate-buffered saline (PBS). L-Ago2 WT (n = 10 for PBS and n = 13 for S961) and KO (n = 11 for PBS and n = 14 for S961) mice fed NCD at 9 weeks of age were continuously treated with S961 (10 nM/week) via osmotic pumps. The graph on the right shows an integrated area under the glucose disposal curves (AUC) for each condition. b Serum insulin levels after daytime food withdrawal for 6 h in L-Ago2 WT and KO mice at 2 weeks post S961 (n = 7, each genotype) or PBS (n = 6, each genotype) treatment. c Hepatic glycogen contents in L-Ago2 WT and KO mice at 2 weeks post S961 (n = 7, each genotype) or PBS (n = 5, each genotype) treatment. d A heatmap diagram illustrating the differential expression of mature miRNAs in the liver of L-Ago2 WT and KO mice treated with PBS or S961 for 2 weeks. Significant miRNAs differentially expressed between genotypes were identified using DESeq2 (|fold change| > 1.25x and adjusted p < 0.0005). Clusters I and IV are miRNAs differentially expressed between L-Ago2 WT (n = 6) and L-Ago2 KO (n = 6) groups. Clusters II and III are miRNAs differentially expressed by S961 treatment in WT and KO groups, respectively. The log2 expression values were scaled by z-score. e, f Effect of S961-treatment on expression of MD-miRNAs (e) and genes regulating energy metabolism (f) in the liver of L-Ago2 WT and KO mice treated with PBS or S961 (n = 6, each group) for 2 weeks. Data are shown as the mean ± SEM. *p < 0.05, **p < 0.01 the insulin antagonist peptide, S961 (43 amino acids in length). hepatic Ago2-deficiency on S961-induced deterioration of glucose S961 binds to the insulin receptor and blocks insulin signaling metabolism. At this age, while glucose metabolism assessed by in vivo to acutely induce hyperglycemia38. This model allowed us GTT was comparable between the genotypes in a PBS-treated to further assess the role of hepatic Ago2 in glycemic control control group, S961 treatment caused hyperglycemia in WT mice without the potential confounding effects of body weight and 1 week after treatment (Fig. 3a), and hyperinsulinemia and gly- adiposity. After infusing S961 into L-Ago2 WT and L-Ago2 KO cogenolysis in WT mice after 2 weeks of treatment (Fig. 3b, c). mice fed NCD at 12 weeks of age, we examined the effect of Remarkably, L-Ago2 KO mice are resistant to S961-induced

NATURE COMMUNICATIONS | (2018)9:3658 | DOI: 10.1038/s41467-018-05870-6 | www.nature.com/naturecommunications 5 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-05870-6

a bcd 500 300 500 50 * ) ) ** ) 250 –1 45 –1 –1 400 400 40 200 35 ** 300 300 30 ** 150 * 25 200 200 20 100 15 Body weight (g) Body weight HFD L-Ago2 WT 100 100 Blood glucose (mg dl Blood glucose (mg dl 50 10 HFD L-Ago2 KO Blood glucose (mg dl L-Ago2 WT_HFD L-Ago2 WT_HFD L-Ago2 WT_HFD L-Ago2 KO_HFD L-Ago2 KO_HFD 5 CD L-Ago2 WT L-Ago2 KO_HFD CD L-Ago2 KO 0 0 0 0 0 30 60 90120 0 30 60 90 120 03060 90 120 4 6 8 1012141618202224 Time after IP glucose (min) Time after IP insulin (min) Time after IP pyruvate (min) Weeks of age

e AUCf Insulin g Gastrocnemius Visceral Subcutaneous h sHGP 30 1.6 25 clearance 16 3 fat 3 fat 70 * * 25 1.4 14 60 20 2.5 2.5 1.2 12

–1 –1 50 –1 –1 20 –1 2 2 1 15 10 min min min 40 min min % –1 –1 –1

15 –1 0.8 –1 8 1.5 1.5 30 0.6 10 6 mg kg mg 10 kg mg

ml kg 1 1 mg kg mg mg kg mg Relative levels Relative 20 0.4 4 L-Ago2 WT_HFD 5 5 0.5 0.5 L-Ago2 KO_HFD 0.2 2 10 0 0 00 0 0 0 –10 20 40 60 80 90 110 L-Ago2 L-Ago2 L-Ago2 L-Ago2 L-Ago2 L-Ago2 L-Ago2 L-Ago2 L-Ago2 L-Ago2 L-Ago2 L-Ago2 min WT KO WT KO WT KO WT KO WT KO WT KO

Fig. 4 Ago2-deficiency in the liver improves glucose metabolism in obesity. a Body weights of L-Ago2 WT (n = 16) and KO (n = 17) mice fed HFD, and L- Ago2 WT (n = 15) and KO (n = 11) mice fed CD, starting in 4 weeks of age. b GTT performed in L-Ago2 WT (n = 16) and KO (n = 17) mice fed HFD at 20 weeks of age. c ITT performed in L-Ago2 WT (n = 16) and KO (n = 17) mice fed HFD at 14 weeks of age. d PTT performed in L-Ago2 WT (n = 8) and KO (n = 8) mice fed HFD at 17 weeks of age. e–h Hyperinsulinemic-euglycemic clamp studies performed in L-Ago2 WT (n = 6) and L-Ago2 KO (n = 10) mice fed HFD for 20 weeks. e Glucose infusion rates (GIR) throughout the clamp procedures. The graph on the right shows an integrated area under curves (AUC) of GIR. f Insulin clearance levels during the clamp. g Tissue glucose uptakes in gastrocnemius muscle, visceral fat, and subcutaneous fat tissues. h Suppression of hepatic glucose production (sHGP) during the clamp. Data are shown as the mean ± SEM. *p < 0.05, **p < 0.01 glucose intolerance after 1 week of treatment (Fig. 3a) and S961 Critical roles of hepatic Ago2 in energy metabolism on high- treatment of L-Ago2 KO mice resulted in a lower induction of fat-diet challenge. We next asked if nutrient challenge might plasma insulin levels and higher hepatic glycogen contents accentuate Ago2’s role in metabolic regulation. We thus compared with control mice (Fig. 3b, c). These data indicate that employed a high-fat diet (HFD)-induced obesity model that inactivation of hepatic Ago2 improves systemic glucose metabo- induces insulin resistance, glucose intolerance, and hepatic stea- lism in the condition of insulin insufficiency. tosis. We placed L-Ago2 KO, and L-Ago1 KO, and control WT Expression profiles that assessed the effect of insulin insuffi- mice on HFD or a control diet (CD), commencing at 4 weeks of ciency on miRNAs in the liver categorized four different classes of age (Fig. 4a and Supplementary Fig. 3a). Of note, the body miRNAs; (I) dominantly expressed in L-Ago2 WT, (II) induced weights of the HFD-fed L-Ago2 KO mice became lower than that by S961 in both L-Ago2 WT and KO, (III) suppressed by S961 in of controls and the difference reached statistical significance at both L-Ago2 WT and KO, and (IV) dominantly expressed in L- 22 weeks of age (Fig. 4a). Conversely, the body weight of L-Ago1 Ago2 KO (Fig. 3d). Importantly, S961 treatment did not KO mice was comparable to controls in the HFD condition modulate expression levels of miRNAs categorized in class (IV) (Supplementary Fig. 3a). The improvements in glucose metabo- in both L-Ago2 WT and KO liver. Conversely, several miRNAs in lism observed in the HFD-fed L-Ago2 KO were even more pro- class (I) that contains MD-miRNAs including miR-802, miR-103/ nounced when compared to the CD-feeding condition, as 107, and miR-130a were strikingly induced by S961 treatment in assessed by GTT, ITT, and PTT (Fig. 4b–d, Supplementary Fig. 4a the WT liver in an Ago2-dependent manner (Fig. 3d, e). These and b). These improvements were not observed in L-Ago1 KO fed analyses implicate hepatic Ago2-dependent miRNAs play a role HFD (Supplementary Fig. 3b–f). Consistent with improved sys- in the disruption of glucose metabolism under the condition of temic glucose tolerance and insulin sensitivity in L-Ago2 KO insulin insufficiency. Despite the decrease of miR-802 and miR- mice, HFD-induced pancreatic β-cell proliferation and islet 103/107 expression in the Ago2-deficient condition, expression hypertrophy were attenuated compared to L-Ago2 WT mice levels of their known targets, Hnf1β and Cav1, were comparable (Supplementary Fig. 4c–g), supporting that hepatic Ago2- between the genotypes even in PBS-treated groups at this age deficiency improves insulin sensitivity in the pathogenesis of (Fig. 3f). While expression of glucose-6-phosphatase (G6Pase), a obesity. gluconeogenic gene, was similarly induced between the genotypes To further demonstrate the role of hepatic Ago2, we performed (Fig. 3f), expressions of genes critical for mitochondrial function, the hyperinsulinemic-euglycemic clamp study to examine the such as Pgc1α, peroxisome proliferator-activated receptor alpha whole-body glucose metabolism and insulin sensitivity. Glucose (Pparα), mitochondrial transcription factor (Tfam), and citrate infusion rates (GIR) during the clamp studies indicated that L- synthase (Cs) were higher in the liver of L-Ago2 KO mice. These Ago2 KO mice required significantly higher levels of glucose results, along with the Ago2-deficient hepatocyte analyses, suggest infusion to maintain blood glucose consistent with increased that Ago2 regulates the program of a subset of miRNAs involved insulin sensitivity (Fig. 4e and Supplementary Fig. 4h). Insulin in energy metabolism, which may be associated with enhance- clearance and glucose uptakes in gastrocnemius muscle, visceral ment of gluconeogenesis and suppression of glycolysis and and subcutaneous fat, brown adipose tissue, and heart were hepatic mitochondrial oxidation induced by insulin insufficiency. comparable (Fig. 4f, g and Supplementary Fig. 4j). Conversely,

6 NATURE COMMUNICATIONS | (2018)9:3658 | DOI: 10.1038/s41467-018-05870-6 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-05870-6 ARTICLE

abcdLiver weight TG content ALT e 3.5 5 * 500 * 3.5 L-Ago2 WT * L-Ago2 KO 3 CD L-Ago2 WT HFD L-Ago2 WT ** 4 400 3 ) –1 2.5 liver) 2.5 –1 2 3 300 ** 100 µm 100 µm 2 1.5 ** ** 2 200 CD L-Ago2 KO HFD L-Ago2 KO 1.5 ** ** **

Tissue weight (g) Tissue weight 1 1 Serum ALT (IU L Serum ALT

1 100 mRNA levels Relative 0.5 TG content (mg g 0.5

100 µm 100 µm 0 0 0 0 CD CD HFD HFD L-Ago2 L-Ago2 L-Ago2 L-Ago2 WT KO WT KO Pdhb Fh1 Fasn WT KO WT KO Cpt1a Acss1ldh3b Mdh1 Scd1 Sod1 Sod2 Ucp2 Srebp1c

f Palmitate oxidation Acetate oxidation g Fat loss Lean loss h mtDNA copy i Energy expenditure 25 15 * ** ** 25 ** 0 0 * Light Dark 1.4 18 * protein) 20 protein) 16 –1 20 1.2 ) –2 ) –1 g –1 –1 g * µ ** ** 14 10 µ 1

–1 kg kg

–1 15 12

15 –1 –5 –4 0.8 –1 10 10 0.6 8 5 –6 0.4 10 6 Relative levels Relative Percent of change Percent 5 0.2 4 Heat (kcal h Heat (kcal h –10 ** –8 0 5 L-Ago2 WT 2 OCR (pMole min

0 OCR (pMole min 0 L-Ago2 KO 0 Palmitate––++ Acetate ––++ L-Ago2 WT L-Ago2 KO L-Ago2 L-Ago2 L-Ago2 L-Ago2 0 WT KO WT KO 0 6 12 18 24 30 36 42 48 54 60 66 L-Ago2L-Ago2 WT L-Ago2 KO L-Ago2 WT KO Time (h)

Fig. 5 Ago2-deficiency in the liver prevents hepatic steatosis with enhanced energy expenditure. a Liver weight in L-Ago2 WT mice fed CD (n = 5), L-Ago2 KO mice fed CD (n = 4), L-Ago2 WT mice fed HFD (n = 5), and L-Ago2 KO mice fed HFD (n = 5) at 30 weeks of age. b, c Liver triglyceride (TG) contents (b) and serum ALT levels (c) in L-Ago2 WT (n = 7) and L-Ago2 KO (n = 5) mice fed HFD at 23 weeks of age. d H&E-stained sections of the liver in each genotype at 30 weeks of age. Scale bar, 100 μm. e Expression levels of key mRNAs involved in energy metabolism in the liver of L-Ago2 WT (n = 8) and L- Ago2 KO (n = 5) mice fed HFD at 23 weeks of age. f Levels of β-oxidation in the presence of 0.12 mM palmitate and mitochondrial OCR in the presence of 5 mM acetate in primary hepatocytes isolated from L-Ago2 WT (n = 3 for control, n = 3 palmitate and acetate, respectively) and L-Ago2 KO mice (n = 3 for control, n = 3 palmitate and acetate, respectively). g Effects of a 14-h fast on fat mass, and lean body mass in L-Ago2 WT (n = 9) and KO (n = 10) mice fed HFD at 20 weeks of age. h Copy numbers of mtDNA in L-Ago2 WT (n = 7) and KO (n = 5) mice fed HFD at 23 weeks of age. i Energy expenditure in L- Ago2 WT (n = 8) and KO (n = 8) mice fed HFD at 16 weeks of age. Data are shown as the mean ± SEM. *p < 0.05, **p < 0.01 hepatic glucose production was significantly suppressed in L- (Supplementary Fig. 5e), while lean mass composition of L-Ago2 Ago2 KO mice (Fig. 4h and Supplementary Fig. 4k). These studies KO mice was slightly higher than that of control mice on HFD at indicate that the liver is the main locus responsible for improving 20 weeks of age (Supplementary Fig. 5f). In addition, the ability to systemic insulin sensitivity and glucose metabolism in L-Ago2 utilize fat mass, which was calculated by measurements of fat and KO mice. lean compositions before and after an overnight fast, was higher Importantly, the liver of L-Ago2 KO mice was characterized by in L-Ago2 KO mice on HFD compared to WT controls (1.675- lowered liver weights and triglyceride content, accompanied by fold higher in L-Ago2 KO: p < 0.01, t-test) (Fig. 5g). In support of lower serum ALT levels on HFD (Fig. 5a–c). In addition, plasma these observations, there was an increased copy number of triglyceride levels were lower in L-Ago2 KO mice fed HFD, while mitochondrial-DNA (mtDNA) in the Ago2-deficient liver in those of cholesterol, phospholipids, and free fatty acids were obesity (Fig. 5h). Consistently, rates of energy expenditure of L- comparable between the genotypes (Supplementary Fig. 5a–d). Ago2 KO mice fed HFD for 12 weeks were significantly higher There was also a reduction in hepatic fatty infiltration as than those of controls (Fig. 5i and Supplementary Fig. 5g and h), visualized by haematoxylin and eosin (H&E) staining (Fig. 5d). despite no significant changes in body weight, total physical While levels of genes involved in lipid biosynthesis such as activity, food intake, or amounts of fecal lipids between genotypes stearoyl-CoA desaturase-1 (Scd1) and fatty acid synthease (Fasn) (Supplementary Fig. 5i–l). We performed similar experiments were comparable between the genotypes, those of carnitine with L-Ago1 WT and L-Ago1 KO mice fed HFD and found that palmitoyltransferase 1A (Cpt1a) and acetyl-coenzyme A synthe- there were no differences in the regulation of energy homeostasis tase 2-like (Acss1) that mediate catabolic processes of fatty acids between the genotypes (Supplementary Fig. 3g–m). Taken were increased in the livers of L-Ago2 KO mice (Fig. 5e). together, these data indicate that inactivation of Ago2, but not Consistently, mitochondrial OCR in response to palmitate and Ago1, in the liver increases mitochondrial capacity and energy acetate whose circulating levels are positively correlated with expenditure, which appears to link to improvement of obesity- obesity and its related sequelae39 was significantly higher in associated pathophysiology. Ago2-deficient hepatocytes compared to controls (Fig. 5f). These cellular phenotypes could promote reduction of hepatic triglycer- Ago2-mediated RNA silencing regulates expression of genes ide accumulation and lowered hepatic steatosis levels we have involved in energy metabolism. To investigate molecular observed in the liver of L-Ago2 KO mice fed HFD. Quantitative mechanisms by which Ago2 orchestrates hepatic energy meta- Magnetic Resonance technology (EchoMRI) analyses revealed bolism, we additionally profiled hepatic miRNA expression under that total body fat mass in L-Ago2 KO mice is lower than in WT the condition of HFD (Fig. 6a). Utilizing the list of significant controls (44.3% fat reduction in L-Ago2 KO: p < 0.05, t-test) miRNAs on HFD, the miRNA target pathway enrichment

NATURE COMMUNICATIONS | (2018)9:3658 | DOI: 10.1038/s41467-018-05870-6 | www.nature.com/naturecommunications 7 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-05870-6

a L-Ago2 L-Ago2 bc 14 WT KO Down-miRNAs Up-miRNAs ** ** mmu-miR-101b-3p Fatty acid biosynthesis mmu-miR-365-3p mmu-miR-148b-3p PI3K-Akt signaling pathway 12 WT_NCD mmu-miR-107-3p Fatty acid metabolism mmu-miR-194-5p L_Ago2 WT_HFD mmu-miR-30c-5p FoxO signaling pathway mmu-miR-378c Hippo signaling pathway mmu-miR-185-5p L_Ago2 KO_HFD mmu-miR-744-5p Lysine degradation 10 mmu-miR-149-5p Ubiquitin mediated proteolysis mmu-miR-378a-3p MAPK signaling pathway mmu-miR-152-3p mmu-miR-221-3p Phosphatidylinositol signaling mmu-miR-802-5p RAS signaling pathway mmu-miR-1839-5p 8 mmu-miR-23b-3p Glycosaminoglycan biosynthesis mmu-miR-122-5p Insulin signaling pathway mmu-miR-16-5p mmu-miR-27b-3p ErbB signaling pathway mmu-miR-103-3p Circadian rhythm mmu-miR-148a-3p 6 mmu-miR-98-5p TGF-beta signaling pathway mmu-miR-22-5p Protein digestion and absorption mmu-miR-146b-5p cGMP-PKG signaling pathway mmu-miR-199b-5p mmu-miR-100-5p Hedgehog signaling pathway mmu-let-7i-5p cAMP signaling pathway miRNA levels Relative 4 mmu-miR-27a-3p mmu-miR-214-3p mRNA suveillance pathway mmu-miR-199a-3p AMPK signaling pathway mmu-miR-199b-3p ** ** ** ** ** ** ** ** ** ** ** ** mmu-miR-125b-5p Protein processing in ER mmu-miR-142a-3p 2-Oxocarboxylic acid metabolism mmu-miR-199a-5p 2 mmu-miR-146a-5p Wnt signaling pathway mmu-miR-10a-5p Inositol phosphate metabolism mmu-miR-342-3p mTOR signaling pathway mmu-miR-10b-5p mmu-miR-181a-5p N-Glycan biosynthesis 0 mmu-miR-143-3p Endocytosis mmu-miR-214-5p mmu-miR-592-5p Sphingolipid signaling pathway Z -score 15 10 5 0 5 10 15 20 –Log10 (p -value) miR-802 miR-107 miR-103 miR-130a miR-148a miR-148b miR-152 –1.5 0 1.5

d e L-Ago2 WT f L-Ago2 WT 1.8 L-Ago2 KO 4 * L-Ago2 WT L-Ago2 KO 1.6 ** 1.2 ** * L-Ago2 KO 3.5 1.4 Luc Control ** 1 3 1.2 ** ** Luc Ampka1 3′ UTR ** * Luc 2.5 ** 1 0.8 ** Ampka1 3′ UTR-M ** miR-148/152 2 * ** ** ** 0.8 0.6 target site mutant 1.5 0.6

Relative levels Relative 0.4 1 0.4 Relative Luc activity Relative Relative mRNA levels Relative 0.5 0.2 0.2 0 0 0 β α Cs Cav1 Hnf1 Pgc1 Ampka1Ampka2 Ampka1 miR-152 ControlAmpka1 UTRAmpka1 miR-148amiR-148b 3′ pri-miR-152 ′ UTR-M pri-miR-148apri-miR-148bpri-miR-152 pri-miR-148apri-miR-148b 3

g h Hepatic Ago2 25 25 40 L-Ago2 WT L-Ago2 WT L-Ago2 WT * Maturation of L-Ago2 KO * L-Ago2 KO 35 L-Ago2 KO 20 20 * MD-related miRNAs 30 15 * 25 15 Ampka1 * mRNA (%) 20 Other targets * mRNA (%) 10 mRNA (%) 10 15 Cs -actin

! 10 Ampka1 5 5 Glucose and lipid 5 metabolsm (oxidation) 0 0 0 12345678 9 101112 1 2 3 4 56 78 9 101112 1 2 3 4 56 78 9 101112 Fraction number Fraction number Fraction number

Fig. 6 Hepatic Ago2 regulates expression of MD-miRNAs and Ampka1 in obesity. a A heatmap diagram illustrating the differential expression of mature miRNAs in the liver of L-Ago2 WT (n = 3) and L-Ago2 KO (n = 3) mice fed HFD for 16 weeks. Significant miRNAs differentially expressed between L-Ago2 WT and L-Ago2 KO groups were identified using DESeq2 (|fold change| > 2x and adjusted p < 0.05) and plotted as a heatmap using z-score. b Metabolic pathway enrichment analysis of miRNAs significantly down-regulated (blue) and up-regulated (red) miRNAs in the liver of L-Ago2 KO mice fed HFD for 16 weeks. These miRNAs were queried to calculate the most enriched KEGG pathways using DIANA-mirPath web-server (p < 0.05 and MicroT threshold < 0.8). Pathways unrelated to liver functions were excluded in this pyramid plot. c Expression levels of specific MD-miRNAs in the liver of L-Ago2 WT mice (n = 8) fed NCD at 25 weeks of age, and L-Ago2 WT (n = 7) and L-Ago2 KO (n = 5) mice fed HFD at 23 weeks of age. d Expression levels of MD-miRNAs’ target mRNAs and pri-miRNAs in the liver of L-Ago2 WT (n = 7) and L-Ago2 KO (n = 5) mice fed HFD at 25 weeks of age. e Compared expression levels of Ampka1, miR-148/152, and their pri-miRNAs in primary hepatocytes isolated from L-Ago2 WT and L-Ago2 KO mice. f Relative luciferase activity by which Ampka1 3' UTR with or without harboring a mutation at miR148/152 putative target site was assessed in primary hepatocytes isolated from L-Ago2 WT (n = 4 for control, n = 6 for Ampka1 3′ UTR and Ampka1 3′ UTR-M, respectively) and L-Ago2 KO mice (n = 4 for control, n = 6 for Ampka1 3′ UTR and Ampka1 3′ UTR-M, respectively). g Quantification of Ampka1, Cs, and β-actin mRNA levels in fractions collected from polysome profiles of primary hepatocytes isolated from L-Ago2 WT and L-Ago2 KO mice. The graphs show the quantification of the results. h A proposed role of hepatic Ago2 in the regulation of glucose and lipid metabolism in the liver. Data are shown as the mean ± SEM. *p < 0.05, **p < 0.01

8 NATURE COMMUNICATIONS | (2018)9:3658 | DOI: 10.1038/s41467-018-05870-6 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-05870-6 ARTICLE analysis revealed that hepatic Ago2 deficiency affected several with miRNA sequencing data obtained under lean, HFD, and functional clades such as glucose and lipid metabolism and S961-treated conditions (Supplementary Fig. 6c and d). With the protein translation regulation (Fig. 6b). While Ago2 protein levels Ago2-dependent miRNAs, we extracted lists of predicted were slightly decreased in the liver of mice fed HFD (Supple- conserved target genes involved in energy metabolism from the mental Fig. 6a), several MD-miRNAs were highly induced in the widely used TargetScan 7.1 website. This analysis identified a liver of L-Ago2 WT mice fed HFD and a leptin-deficient (ob/ob) subset of genes, including Ampka1 (also known as Prkaa1,a obese mice (Fig. 6c and Supplemental Fig. 6b). Importantly, these catalytic subunit of AMPK that plays a critical role in AMPK miRNAs are constantly decreased in that of L-Ago2 KO mice activation) and Cs, that have 3′ untranslated region (UTR) (Fig. 6a, c). Consistent with these observations, expression levels containing multiple target sites for Ago2-dependent miRNAs of their known target mRNAs, such as Hnf1β, Cav1, and Pgc1α, including miR-148a/152 known to evoke hyperlipidemia, are increased in L-Ago2 KO liver (Fig. 6d). Of note, expression hypercholesteremia, and atherosclerosis14,15 (Supplementary levels of these genes were comparable between L-Ago1 WT and Fig. 6e and f). As AMPK is known as a critical regulator KO liver (Supplementary Fig. 3n). of energy metabolism, we further assessed the role of Ago2 To explore targets of Ago2-dependent MD-miRNAs for in Ampka1 expression. Analyzing a public database of metabolic regulation, we then took a bioinformatics approach photoactivatable ribonucleoside-enhanced crosslinking and

a L-Ago2 WT b L-Ago2 WT 4 L-Ago2 WT L-Ago2 KO 2.5 L-Ago2 KO L-Ago2 KO * 3.5 ** AMPKα1 2 * 3 p-AMPKα * 1.5 2.5 AMPKα 2 ** ** ** ** ** ** ** p-ACC 1 1.5 ** * *

Relative levels Relative 1 ACC 0.5

Relative mRNA levels Relative 0.5 S6 0 0 1 α α Nd1 Nd2 Nd3 Nd6 Tfam Cox1Cox2 Cox3 Cytb Nd4Nd4L Nd5 Atp6 Atp8 p-ACC AMPK p-AMPK

c d ef L-Ago2 WT L-Ago2 KO L-Ago2 WT L-Ago KO Albumin 1.6 Ago2 WT 1.2 2 50 Ago2 KO * 2.5 * ) * 1.4 1.8 –1 * 1 * 1.6 2 40 1.2 * 0.8 * 1.4 ** * * 1 1.2 * 1.5 30 1 0.8 0.6 0.8 1 20 0.6 0.4

0.6 levels Relative

Relative levels Relative 0.4 0.4 0.5 10 0.2 Ago2 WT Reduction of ATP/ADP 0.2 protein levels Relative 0.2 Serum albumin (mg ml Serum albumin Ago2 KO 0 0 0 0 0 α Cs L-Ago2 L-Ago2 05 052 ATP ADP Total Pgc1 WT KO Met treatment (h) Met treatment (h) Ampka1 Albumin -Tubulin ATP/ADP protein β

g h 1.4 i p-AMPKα/AMPKα ContPhen Cont Rote ** 7 1.2 Hepatic Ago2 * L-Ago2: WT KOWT KO WT KO WT KO 6 1 L-Ago2 WT L-Ago2 KO * 5 0.8 mRNA 025025 Met (h) translation 4 0.6 p-AMPKα 3 Relative levels Relative 0.4 AMPKα 2 Energy Relative levels Relative 0.2 consumption β-Actin 1 0 0 0 25025 hhhhhh AMPK activation L-Ago2 L-Ago2 L-Ago2 KO L-Ago2 KO L-Ago2 KO L-Ago2 WT L-Ago2 WT L-Ago2 WT L-Ago2 WT KO CBB Nascent protein synthesis Cont Phen Rote

Fig. 7 Hepatic Ago2-deficiency enhances energy expenditure associated with protein synthesis and AMPK activation. a, b Western blot analyses of AMPK expression and activation (a) and mRNA expression of the Tfam-mitochondrial genes (b) in the liver of L-Ago2 WT (n = 5) and L-Ago2 KO (n = 5) fed HFD at 25 weeks of age. c ATP, ADP, and ATP/ADP ratio levels in L-Ago2 WT (n = 5) and L-Ago2 KO (n = 5) mice fed HFD at 25 weeks of age. ATP/ADP ratio levels were independently measured with a distinct procedure from the ATP and ADP assays. d Western blot analysis of total and specific protein levels normalized by 12S-genomic DNA in the liver of L-Ago2 WT (n = 5) and KO (n = 5) mice fed HFD at 30 weeks of age. e Serum albumin levels in L- Ago2 WT (n = 8) and L-Ago2 KO (n = 8) mice fed HFD at 25 weeks of age. f Energy consumption rate measured in primary hepatocytes isolated from L- Ago2 WT (n = 4, n = 4, n = 3 for 0, 2 h, 5 h, respectively) and L-Ago2 KO (n = 4, n = 4, n = 4 for 0, 2 h, 5 h, respectively) mice in the presence of 1 mM metformin. g Effect of Ago2-deficiency on expression of AMPK activation in primary hepatocytes isolated from L-Ago2 WT (n = 8, n = 8, n = 8 for 0, 2 h, 5 h, respectively) and L-Ago2 KO (n = 8, n = 8, n = 8 for 0, 2, 5 h, respectively) mice in the presence of 1 mM metformin. The graphs show the quantification of the results. h Effect of Ago2 on nascent protein synthesis. Primary hepatocytes isolated from L-Ago2 WT and L-Ago2 KO mice were treated with or without 200 μM Phenformin (Phen) or 10 μM Rotenone (Rote) for 5 h (n = 15 for control, n = 9 for Phen, n = 6 for Rote). i A proposed role of hepatic Ago2 in suppression of protein translation, leading to AMPK activation. The graphs show the quantification of the results. Data are shown as the mean ± SEM. *p < 0.05, **p < 0.01

NATURE COMMUNICATIONS | (2018)9:3658 | DOI: 10.1038/s41467-018-05870-6 | www.nature.com/naturecommunications 9 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-05870-6 immunoprecipitation (PAR-CLIP) with Ago240 revealed that suggest that enhanced protein synthesis in the liver may result in Ago2 binds to the region of the Ampka1 3′ UTR that contains a lowered ATP/ADP ratio and AMPK activation in L-Ago2 KO binding sites of Ago2-dependent miRNAs including miR-148/152 mice. and miR-130a (Supplementary Fig. 6g). Consistent with these To further examine if Ago2 deficiency accelerates cellular findings, we confirmed the induction of Ampka1 expression, energy consumption associated with protein synthesis, we treated accompanied by the reduction of miR-148a, miR-148b, and miR- primary hepatocytes with metformin, which is an anti-diabetic 152, in the liver of L-Ago2 KO mice fed HFD and in Ago2- drug and inhibits mitochondrial respiratory-chain complex I deficient primary hepatocytes (Fig. 6d, f). We next conducted activity restricting ATP generation41, and measured ATP/ADP luciferase assays in which Ampka1’s 3′ UTR, with or without levels. Consistent with enhanced capacity for energy production harboring a mutation at the miR-148/152 target site, was sub- in Ago2-deficiency (Fig. 2g), relative ATP/ADP levels were higher cloned into luciferase expression vector. The luciferase activity of in Ago2-deficient hepatocytes compared to controls, however, the each was measured in primary hepatocytes isolated from L-Ago2 levels were rapidly decreased post-metformin treatment, suggest- WT and KO mice. Luciferase activity was higher in Ago2 KO ing a higher energy consumption rate in Ago2-deficiency (Fig. 7f). hepatocyte transfected with Ampkα13′ UTR without a mutation Consistently, levels of metformin-induced AMPK activation in compared with Ago2 WT hepatocyte, but the induction in Ago2 Ago2-deficient hepatocytes were significantly higher than that in KO hepatocyte disappeared in the setting with mutated Ampka1 controls (Fig. 7g). To directly investigate the effect of Ago2- 3' UTR (Fig. 6f). These analyses demonstrated that miR-148/152 deficiency on protein synthesis in hepatocytes, we measured the are involved in suppression of Ampka1 expression in a manner levels of nascent protein synthesis. Compared to WT controls, the dependent on Ago2 and a miR-148/152 target site (Fig. 6f). As levels were significantly increased in Ago2-deficient hepatocytes miRNA inhibits the translation of target mRNAs through RNA (Fig. 7h). By restricting energy supply using phenformin and silencing, we additionally asked if Ago2-deficiency affects rotenone, both of which inhibit mitochondrial respiratory-chain translation of genes having target sites of MD-miRNAs by complex I activity, the levels of nascent protein synthesis in Ago2- investigating polysome-bound mRNA expression patterns. deficient hepatocytes were equivalent or still higher compared to Expression levels of polysome-bound Ampka1 and Cs were those in controls (Fig. 7g). enriched in Ago2-deficient primary hepatocytes, while those of β- Since Ago2's slicer activity uniquely regulates RNA silencing, Actin were comparable (Fig. 6g). Taken together, these findings we then asked if the slicer activity is involved in the regulation of further confirm that hepatic Ago2-mediated MD-miRNA expres- energy consumption and AMPK activation. Ago2-deficient MEFs sion and RNA silencing are linked to expressional regulation of were characterized by enhanced expression of Ampka1, and genes involved in energy metabolism (Fig. 6h). reconstitution of the MEFs with WT Ago2 suppressed expression of both Ampka1 mRNA and AMPKα protein, whereas the Ago2 D669A mutant did not (Supplementary Fig. 8a). We also Hepatic Ago2 regulates energy consumption associated with observed that AMPK activity, assessed by phosphorylation levels AMPK activation. We next examined AMPKα1 protein levels of AMPKα and ACC, was higher in Ago2-deficient MEFs under and noticed that Ago2 deficiency increased not only the protein serum starvation condition (Supplementary Fig. 8a). While levels but also activity of AMPK, assessed by phosphorylation activated AMPKα is known to suppress mRNA translation42, levels of AMPKα, AMPKβ, and an AMPK substrate, Acetyl-CoA levels of nascent protein synthesis in Ago2-deficient MEFs are carboxylase (ACC), in the liver of L-Ago2 KO mice fed HFD and increased compared to the cells reconstituted with WT Ago2 treated with S961 (Fig. 7a and Supplementary Fig. 7a and b). In (Supplementary Fig. 8b). To investigate if enhanced protein agreement with the activation of AMPK, other AMPK substrates, synthesis reasons AMPK activation in Ago2 deficiency, we treated UNC-51-like kinase 1 (ULK1), and Mitochondrial fission factor MEFs with cycloheximide (CHX), a protein synthesis inhibitor, (MFF) are increased in the liver of L-Ago2 KO mice (Supple- and monitored AMPK activation. Inhibition of protein synthesis mentary Fig. 7a), suggesting enhanced autophagy/mitophagy and resulted in reduction of AMPK activation in WT MEFs, and the improved mitochondrial quality in the Ago2-deficient liver in effect became more robust in Ago2-deficient MEFs, indicating obesity. Consistently, expression levels of the Tfam-mitochondrial that Ago2 suppresses protein synthesis-mediated energy con- gene pathway are increased in the liver of L-Ago2 KO mice fed sumption (Supplementary Fig. 8c). Taken together, these results HFD (Fig. 7b). indicate that, in addition to an increase of Ampka1 expression, AMPK is activated when cellular energy level becomes low. there is enhanced energy consumption associated with protein Indeed, we found a profound induction of ADP levels in the synthesis, leading to the lowered ATP/ADP ratio, which appears Ago2-deficient liver, while ATP levels are comparable between to enhance AMPK activation in Ago2-deficient conditions the genotypes, leading to a reduction of ATP/ADP ratio in the (Fig. 7i). liver of L-Ago2 KO mice fed HFD (Fig. 7c). While Ago2- deficiency enhances capacity for mitochondrial oxidation and ATP production in hepatocytes (Figs. 2g, 5f), systemic energy Hepatic Ago2 deficiency improves glucose metabolism in a expenditure is also increased (Fig. 5i). Therefore, we hypothesized hepatic Ampka1-deficient condition. While hepatic Ago2- that both energy production and consumption are enhanced in deficiency reduces expression of a specific repertoire of MD- the liver of L-Ago2 KO mice compared to their controls. Since a miRNAs of which some of them target Ampka1, it is obvious that main function of Ago2 is to suppress protein translation, which is changes in expression of these miRNAs also affects translation of one of the most energy consuming cellular processes, through other target mRNAs. Similarly, enhanced protein synthesis in the RNA silencing, we investigated the effect of Ago2-deficiency on liver of L-Ago2 KO mice must influence not only AMPK acti- protein synthesis in the liver. Levels of total and specific proteins vation but also other cellular events linked metabolic regulation. normalized by DNA contents were higher in the liver of L-Ago2 To clarify the role of Ampka1 in the metabolic alterations in L- KO mice (Fig. 7d and Supplementary Fig. 7c). Similarly, Ago2 KO mice, we generated liver-specific Ampka1- and Ago2- examination of the levels of hepatic and serum albumin, one of deficient mice (L-DKO mice) and placed them and their control the most abundant circulating proteins produced by the liver, groups, L-Ampka1 WT and L-Ampka1 KO mice, on HFD for revealed that the albumin levels were increased in L-Ago2 KO analyses of glucose metabolism. While no significant difference mice compared to their controls (Fig. 7d, e). These observations was observed in body weight and fasting blood glucose levels

10 NATURE COMMUNICATIONS | (2018)9:3658 | DOI: 10.1038/s41467-018-05870-6 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-05870-6 ARTICLE

abcdef 500 AUC Body weight Blood glucose Insulin HOMA IR Insulin sensitivity 1.2 30 250 250 5 ** 80 ) ** * * * –1 400 ) 4.5 70 * 1 * 25 –1 200 200 4 ) 60 –1 3.5 300 0.8 20 150 150 3 50 0.6 15 2.5 40

200 Units 100 100 2 30 0.4 10 1.5 Relative levels Relative

L-Ampka1 WT (g) Body weight 100 Insulin (pmol L 20

Blood glucose (mg dL L-Ampka1 KO 50 50 1

0.2 5 Insulin sensitivity (%S) Blood glucose (mg dL 10 L-DKO 0.5 0 0 0 0 0 0 0 0 15 30 60 90 120 Time after IP glucose (min) L-Ampka1 WT L-Ampka1 KO L-DKO

g h Hepatic ago2 deficiency 7 ** 6 * L-Ampka1 WT L-Ampka1 KO L-DKO MD-miRNA biogenesis General 5 and RNA silencing RNA silencing 4 ** ** 3 * * ** Glucose and lipid mRNA ** ** ** metabolism translation Relative levels Relative ** * 2 **** ** ** ** ** ** ** ** ** * ** *** * ** * ** ** 1 ** ** ** * ** ** ATP ATP 0 production consumption β α α Cs Tfam Enhanced flow of cellular energy Hnf1 Cav1Pgc1 Ppar miR-802miR-107miR-103miR-130amiR-148amiR-148bmiR-152 pri-miR-802pri-miR-107 pri-miR-148a pri-miR-152 pri-miR-103-1pri-miR-103-2pri-miR-103a pri-miR-148b Improved metabolism in obesity

Fig. 8 Effects of hepatic Ago2-deficiency on glucose metabolism in liver-specific Ampka1-deficient mice. a GTT performed in L-Ampka1 WT (n = 17), L-Ampka1 KO (n = 14), and L-DKO (n = 6) mice fed HFD for 5 weeks. b–f Body weight (b), levels of blood glucose (c) and plasma insulin (d), and calculated HOMA IR (e) and insulin sensitivity (f) in L-Ampka1 WT (n = 10), L-Ampka1 KO (n = 8), and L-DKO (n = 6) mice fed HFD for 6 weeks. g Expression levels of miRNAs, pri-miRNAs, and miRNAs in the liver of L-Ampka1 WT (n = 8), L-Ampka1 KO (n = 8), and L-DKO (n = 3) mice fed HFD for 8 weeks. Data are shown as the mean ± SEM. *p < 0.05, **p < 0.01. h A proposed molecular mechanism by which hepatic Ago2-deficiency improves energy metabolism in the pathogenesis of obesity. Ago2-mediated miRNA expression and RNA silencing coordinately suppress mitochondrial oxidation and protein translation, which result in lowered energy supply and consumption. Conversely, hepatic Ago2-deficiency enhances generation of energy from glucose and lipid for protein synthesis, leading to higher ADP and AMP amounts. As a result, the AMPK pathway is activated, leading to improvement of mitochondrial functions and metabolism among these three groups, L-DKO mice exhibited enhanced expression and general mRNA translation. Changes in hepatic glucose tolerance in the condition of HFD feeding for 5 weeks Ago2-mediated energy balance between energy production and (Fig. 8a–c). Plasma insulin levels of L-Ampka1 KO mice were consumption in response to nutrient challenges appears to con- higher than those in L-Ampka1 WT mice and the levels tribute to the pathogenesis of obesity-associated sequelae were drastically decreased in L-DKO mice on HFD (Fig. 8d). (Fig. 8h). These results indicate that inactivation of hepatic Ago2 can Although Ago1 and Ago2 share functional similarities in RNA improve glucose metabolism even in an Ampka1-deficient con- silencing, our study provides evidence that Ago2 has a distinct dition where insulin resistance occurs (Fig. 8d–f). Consistently, role in metabolic regulation. Hepatic Ago1 is dispensable for expression levels of a specific repertoire of MD-miRNAs were obesity-induced pathophysiology, as deletion of hepatic Ago1 did constantly decreased in the liver of L-DKO mice, while levels of not affect diet-induced weight gain, glucose tolerance, or insulin their targets, Hnf1β, Cav1, and Pgc1α, and genes critical for sensitivity. This, in turn, highlights one of the unique functions of enhancing mitochondrial function were higher in the liver of Ago2 in regulating the specific miRNA expression and mRNA L-DKO mice compared to L-Ampka1 WT or L-Ampka1 KO mice silencing. While Ago2-deficiency in the liver affects expression of (Fig. 8g). Taken together, these results suggest that the effect of a small proportion of miRNAs, we demonstrated that Ago2 play a hepatic Ago2-deficiency on glucose metabolism overrides that of critical role in expression of a subset of specific miRNAs, AMPKα1 functions and that a cellular condition which leads to including miR-802, miR-103/107, and miR-148a/152, which are activation of AMPK, but not a direct activation per se, is an known to negatively impact glucose and lipid metabolism. In important factor that improves glucose metabolism observed in addition, expression of these miRNAs is enhanced in response to L-Ago2 KO mice. the energy stress conditions of lower insulin availability or sensitivity, in an Ago2-dependent manner. Under these stress conditions, hepatocytes are normally programmed to stimulate Discussion glucose production, triglyceride synthesis, and the assembly and The role of RNA silencing in suppressing mRNA translation gives secretion of very low-density lipoprotein particles. Hepatic Ago2 rise to the intriguing hypothesis that the RNA silencing is likely integrated into this program through the generation of machinery might be tightly integrated with the regulation of basal selective miRNAs, and by mediating subsequent RNA silencing. metabolic activity and energy homeostasis, as mRNA translation While this mechanism may be beneficial for the maintenance of requires a massive amount of energy. In this study, we uncovered fi systemic energy homeostasis during hypoglycemia, hypermotility, that hepatic Ago2 regulates expression of speci c miRNAs that starvation, and developmental processes, it could also accelerate silence genes critical for glucose and lipid metabolism and the development of metabolic diseases in excess nutrient condi- reduces mRNA translation. These Ago2’s functions appear to tions. When each of the four Ago proteins are ablated con- intimately be linked to energy metabolism by balancing energy stitutively in mice, only the loss of Ago2 causes embryonic production and consumption through regulating specific miRNA

NATURE COMMUNICATIONS | (2018)9:3658 | DOI: 10.1038/s41467-018-05870-6 | www.nature.com/naturecommunications 11 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-05870-6 lethality, whereas loss of other three Ago proteins is dispensable intraperitoneal sodium pyruvate injection (Sigma-Aldrich, 2 g kg−1) following an for animal development24,43–46. Importantly, Ago2’s slicer activ- overnight food withdrawal for 16 h. Body composition was analyzed by EchoMRI™- 53 ity is required for embryonic and perinatal development21. It is 100H instrument (Echo Medical Systems) . To measure body composition after fasting, food was removed from mice for 16 h. To analyze fecal lipid excretion, lipid reasonable to consider a model where Ago2’s unique function content of feces was extracted using chloroform:methanol (2:1) and air-dried under regulates energy metabolism not only in the liver but also in other a fume hood. Mouse serum albumin levels were measured using an ELISA kit organs during development and in adulthood. This may, at least (Abcam). Mouse plasma insulin levels were measured using Mouse Ultrasensitive fi in part, explain the universal importance of Ago2 in such a Insulin ELISA kit (ALPCO). Lipid pro ling was also performed by University of Cincinnati’s MMPC Core. Energy expenditure was measured by using Pheno- diverse array of mammalian organs. Master (TSE Systems)54. Hyperinsulinemic-euglycemic clamp studies were per- In this study, we demonstrated that Ago2-mediated RNA formed at University of Michigan Animal Phenotyping Core. silencing connects the regulation of energy supply with protein biosynthesis. This mechanism may be the core of a vicious cycle Biochemical reagents and antibodies. All biochemical reagents were purchased in disrupted energy metabolism in the obese liver. In this setting, from Sigma-Aldrich unless otherwise indicated. Antibodies against, JNK1 (SC- despite a robust activation of the mTORC1 pathway, protein 1648, 1:2500), Dicer (SC-30226, 1:2500), Akt (SC-8312, 1:2500), phospho-Akt biosynthesis is progressively suppressed, which is a paradox of (Ser473) (SC-7985-R, 1:2500), PGC-1α (SC-13067, 1:2500), β-actin (SC-130656, mRNA translation6,47. While obesity is traditionally considered a 1:5000), and β-tubulin (SC-9104, 1:5000) were from Santa Cruz Biotechnology. state of over-nutrition, recent studies suggest that the obese liver Anti-Acc (3662, 1:2500), anti-phospho-Acc (Ser79) (11818, 1:2500), Anti-AMPKα (5832, 1:2500), Anti-phospho-AMPKα (Thr172) (2535, 1:2500), anti-AMPKβ may, in some aspects, resemble a condition of energy deprivation (4250, 1:2500), anti-phospho-AMPKβ (Ser108) (4181, 1:2500), anti-phospho- in which proper catabolic processes are impaired due to the ULK1 (Ser555) (5869, 1:2500), anti-phospho-ULK1 (Ser317) (12753, 1:2500), anti- repression of oxidative phosphorylation pathways and mito- ULK1 (8054, 1:2500), anti-phospho-MFF (Ser146) (49281, 1:2500), anti-MFF chondrial gene expression7,48. Consistently, obesity is also known (86668, 1:2500), anti-Ago2 (2897, 1:2500), anti-Ago1 (5053, 1:2500), anti-S6 Ribosomal Protein (2217, 1:2500) and anti-phospho-JNK (Thr183/Tyr185) to induce defects in autophagy in the liver, which leads to poor (1:2500), anti-Albumin (4929, 1:2500), anti-Citrate Synthase (14309, 1:2500), anti- 49–51 mitochondrial quality control . As a result, protein bio- AMPKα1 (2795, 1:2500) were purchased from Cell Signaling Technology. Anti- synthesis may be impaired due to under-powered energy supply IRS1 (1:2500) and anti-phospho-IRS (Ser307) (07247, 1:2500) antibodies were even during the activation of the mTORC1 pathway, leading to purchased from Upstate Biotechnology. Anti-AMPKα1 (32047, 1:2500) was pur- further accumulation of energy sources. Conversely, hepatic chased from Abcam. Ago2-deficiency increases expression of key metabolic genes including Ampka1 with enhanced cellular energy consumption Primary hepatocytes. Hepatocytes were isolated from liver of 12–14 weeks old that can lead to lower ATP/ADP ratio. This condition can L-Ago2 WT and L-Ago2 KO mice by a two-step perfusion method55 with a slight fi fl fi amplify activation of AMPK and its substrates ULK1, MFF, and modi cation. Brie y, the liver was rst perfused with 30 ml of HBSS supplemented with 10 mM HEPES, 0.5 mM EGTA and 5 mM glucose and then digested with Pgc1α, leading to improved mitochondrial capacity and quality, 35 ml of Collagenase X (WAKO) at 100 U ml−1 dissolved in HBSS buffer sup- fi which in turn generates suf cient energy for protein biosynthesis. plemented with 10 mM HEPES and 5 mM CaCl2. Liver was collected after perfu- Of note, Ago2 is also known to regulate mRNA silencing through sion and hepatocyte were released and sedimented at 60 × g for 2 min. Hepatocyte interacting with exonuclease complexes, the Ccr4-Not and Pan2- suspension was then layered on a 40% percoll solution (GE Healthcare Life Sci- 52 ences) and centrifuged at 800 × g for 10 min. The alive hepatocytes were recovered Pan3 complexes , in a miRNA-independent manner, which from the bottom of the tube and seeded on culture plates. likely contributes to suppression of protein translation and energy metabolism in the liver. These Ago2-mediated molecular events may solve the paradox of protein biosynthesis in the obese liver, Mouse embryonic fibroblasts. Ago2-deficient fibroblasts reconstituted with Ago2 WT or DA mutant that were kindly provided by Dr. Eric Lai22. We also generated demonstrate a new mechanism in the regulation of basal meta- Ago2fl/fl MEFs through the 3T3 protocol and performed an adenovirus-mediated bolic activity, and provide a novel therapeutic target for metabolic gene transfer for Cre or LacZ expression to obtain Ago2-deficient MEFs56. MEF diseases. cells were cultured in Dulbecco Modified Eagle Medium (DMEM) (Thermo Fisher In conclusion, our results highlight that Ago2 uniquely reg- Scientific: #11965) supplemented with 10% FBS. For western blot analyses of 5 ulates energy production and consumption in the liver, and AMPK, MEF cells were plated at a density of 1 ×10 cells per well of six-well plate for overnight. Next day, cells were kept under serum starvation condition in glu- suggest hepatic Ago2-mediated RNA silencing is a key regulator cose-, pyruvate-, and glutamine-free DEME (Thermo Fisher Scientific: of glucose and lipid metabolism during the pathogenesis of #A1443001). obesity. We also posit that there may be important translational implications for our findings, especially in the design of ther- Quantitative real-time PCR analysis. For mRNA quantification, total RNA was apeutic interventions, to target modulation of a spectrum of extracted using Trizol reagent (Invitrogen). Total RNA was converted to first Ago2-dependent miRNA-mediated events, in chronic metabolic strand cDNA using SuperScript VILO™ cDNA Synthesis Kit (Invitrogen). Quan- disorders, such as diabetes, fatty liver diseases, and other obesity- titative real-time PCR analysis was performed using SYBR Select Master Mix associated sequelae. (Applied Biosystems) in a real-time PCR machine (QuantStudio 6 Flex Real-Time PCR system; Thermo Fisher Scientific). Primers are listed in Supplementary Table 4. To normalize expression data, β-actin mRNA was used as a housekeeping Methods gene. Mice. Animal care and experimental procedures were performed according to For miRNA quantification, total RNA was extracted using miRNeasy Micro Kit procedures approved by the animal care committees of Cincinnati Children’s (Qiagen) according to manufacturer’s instructions. TaqMan miRNA assays (Life Hospital Medical Center. Ago1fl/fl, Ago2fl/fl, and Albumincre/cre were obtained from Technologies) were used and real-time PCR were carried out for mature miRNA the Jackson Laboratory (Stock No: 019001, 016520, and 003574, respectively). quantification. Primary miRNAs were quantified using TaqMan Pri-miRNA Ampka1fl/f mice were kindly provided by Dr. Basilia Zingarelli. All mice used in assays. Sno202 and β-actin were used as internal controls. this study were on C57BL/6 background. Mice were placed on a high-fat diet (HFD: 60% fat, 20% protein, and 20% carbohydrate kcal; Research Diets #D12492) for a diet induced obesity model, a control diet (CD: 10% fat, 20% protein, and 70% High-throughput sequencing of miRNA. Liver tissues were excised from mice, carbohydrate; Research Diets #D12450), or normal chow diet (NCD: 29% Protein, and stored in −80 °C after RNAlater (Invitrogen) treatment. Liver tissue was 13% Fat and 58% Carbohydrate kcal; LAB Diet #5010) beginning at 4 weeks of age homogenized with QIAzol (Qiagen). Total RNA, including miRNA, was extracted ad libitum with free access to water. For acute insulin-resistant model, S961, an using the miRNeasy Micro Kit (Qiagen). High-throughput sequencing of miRNA insulin receptor antagonist, was kindly provided by Dr. Lauge Schaffer38. The was processed according to TruSeq Small RNA Sample Preparation Guide (Illu- ALZET osmotic pump were used to deliver 10 nM S961 or vehicle (PBS) in a mina). Briefly, the total RNA was run on an agarose gel and the band corre- 2 weeks period. GTTs were performed by intraperitoneal glucose injection (1.5 g kg sponding to the size of miRNAs was cut out for further processing. Sequencing −1) following an overnight food withdrawal for 14 h. ITTs were performed by adapters were ligated to the size-selected RNA molecules, followed by reverse intraperitoneal insulin injection (0.75 IU kg−1 for lean mice, 1 IU kg−1 for obese transcription to obtain the cDNA library, which was subsequently sequenced by mice) following a daytime food withdrawal for 6 h. PTTs were performed by Illumina HiSeq2500.

12 NATURE COMMUNICATIONS | (2018)9:3658 | DOI: 10.1038/s41467-018-05870-6 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-05870-6 ARTICLE

Bioinformatic analysis of miRNA seq. Adapter sequences were removed with exogenous fuel substrate supplementation (Seahorse Bioscience) at 37 °C for 1 h. fastx_toolkit (v0.0.14, -a TGGAATTCTCGGGTGCCAAGG -l 15 -M 20 -c, http:// Sequential injection of 2 mM pyruvate, 2 μM oligomycin, 2 μM phenylhydrazone, hannonlab.cshl.edu/fastx_toolkit/) for NCD (Fig. 1a) and S961 (Fig. 3d) samples, and 1 μM rotenone/1 μM antimycin A1 were used to examine mitochondrial and TrimGalore (v0.4.3, -e 0.1 -q 20 -O 1 -a AGATCGGAAGAGCACACGTC, oxidative status. https://www.bioinformatics.babraham.ac.uk/projects/trim_galore/), and Cutadapt For glycolysis stress test, sequential injection of 10 mM glucose, 2 μM (v1.9.1, http://cutadapt.readthedocs.io) for HFD samples (Fig. 6a), with reads that oligomycin, and 2-DG (2-deoxy-glycose, a glucose analog) were used to examine became shorter than 15 bases, remaining reads were filtered with the FASTX glycolysis stress. Each readout was normalized to total proteins. package (v0.0.14), using a quality threshold of 20 over at least 90% of the read. To get raw counts for each mature miRNA, miRExpress (v2.1.4, http://mirexpress. ATP/ADP ratio assay. The ATP/ADP ratio in the mouse liver extract or primary mbc.nctu.edu.tw/) was used to align and create expression count profiles with hepatocytes was determined using a bioluminescent ATP/ADP Ratio Assay Kit default parameters. The alignment step (alignmentSIMD) uses a miRNA precursor (Abcam) according to the manufacturer’s instructions. For mouse liver samples, file (mmu_precursor.txt), supplied with the miRExpress v2.1.4 and the analysis step the samples were immediately frozen in liquid nitrogen and powdered with a uses mmu_miRNA.txt with mature miRNA information for precursor sequences. mortar. Tissue powders were suspended in the provided lysis buffer (10 μl mg−1 of Raw counts were normalized using DESeq2 (v1.18.0, http://bioconductor.org/ tissue powder) for 5 min at room temperature, followed by centrifugation at packages/release/bioc/html/DESeq2.html), to compare mature miRNAs’ relative 10,000 × g for 1 min to pellet insoluble material. For mouse primary hepatocytes, abundance. Differentially expressed miRNAs were predicted also using DESeq2. cells were plated at a density of 0.8 ×105 cells per well of 24-well plate for over- For NCD (Fig. 1a) and HFD samples (Fig. 6a), we applied |fold change| > 2x and night. Next day, cells were cultured with XF base media or DMEM no glucose and adjusted p (FDR BH) < 0.05 as filters. For S961 (Fig. 3d), we applied |fold change| > glutamine media (Gibco: A1443001) in the presence or absence of 5 mM Sodium 1.25x and p < 0.0005. Heatmaps were created by the pheatmap package (https:// Palmitate without FBS for 2 h. Data were normalized by the amount of protein cran.r-project.org/web/packages/pheatmap/index.html). present in the supernatant. miRNA target pathway enrichment analysis. To predict the enriched target ATP and ADP assays. The ATP or ADP in the liver extract was determined using pathways, we used the mirPath web-server (v3, http://snf-515788.vm.okeanos. a ATP Assay Kit (Abcam) or ADP Assay Kit (Abcam), respectively, according to grnet.gr), based on DIANA-microT-CDS algorithm. We chose KEGG (http://www. the manufacturer’s instructions. Data were normalized by the amount of protein genome.jp/kegg) database as a reference and p < 0.05 and MicroT threshold < 0.8 as present in the supernatant. filters to get significantly enriched KEGG pathways.

Protein synthesis analysis. Click-iT labeling technology was used for the detec- Protein extraction and immunoblot analysis. To prepare protein lysates, cells tion of nascent protein synthesis in cells according to manufacturer’s instructions were washed with cold PBS, followed by lysis in cold mammalian cell lysis buffer (Thermo Fisher Scientific). Mouse primary hepatocytes were seeded at 0.6 ×106 [MCLB: 50 mM Tris-HCl (pH 7.4), 150 mM NaCl, 10 mM NaF, 5% glycerol, 1% cells per well in a six-well plate. Cells were then incubated in methionine- and NP-40, 1% protease and phosphatase inhibitor cocktail]. After homogenization on cysteine-free DMEM containing 25 μM of azide-linked methionine analog AHA in ice, the cell lysates were centrifuged, and the supernatants were used for western the presence or absence of 200 μM phenformin or 10 μM Rotenone for 5 h. Azide- blot analyses. For preparation of liver tissue lysates, the tissues were placed in a cold labeled protein lysate from harvested cells was determined by using Click-iT® MCLB and homogenized on ice. The tissue lysates were centrifuged, and the TAMRA Protein Analysis Kit according to manufacturer’s instructions (Thermo supernatants were used for further experiments. Uncropped scans of immunoblots Fisher Scientific) and Typhoon FLA9500 scanner (GE Healthcare) with the exci- are shown in Supplementary Fig. 9. tation at 532 nm. Coomassie Brilliant Blue (CBB)-based staining (Thermo Fisher Scientific; GelCode Blue) for total protein served as a loading control. Morphological and immunohistochemical analysis of hepatic and pancreatic tissues. Liver and pancreas were taken, fixed with 10% formalin, and paraffin- Mitochondrial DNA copy number. Total DNA were purified from mouse liver embedded sections were prepared for further analysis. Paraffin sections were using GeneJet Genomic DNA purification kit according to the manufacturer’s stained with H&E and Periodic Acid-Schiff (PAS) for morphology analyses. For the instruction (Thermo Fisher Scientific). Mitochondrial DNA copy number was immunohistochemical staining, the following primary antibodies were used: guinea detected by qPCR57. pig polyclonal anti-insulin (Abcam) and rabbit monoclonal anti-glucagon (Abcam). The following secondary antibodies were used: Alexa Fluor 488–conjugated AffiniPure Goat Anti-Guinea Pig (Jackson Immunoresearch) and Glucose production assay. Mouse primary hepatocytes were cultured in 12-well Alexa Fluor 594–conjugated AffiniPure Goat Anti-rabbit IgG (Jackson Immu- plates (0.4 ×106 cells per well) in William E supplemented with 10% FBS. Next day, noresearch). The nuclei were stained using 4′,6-diamidino-2-phenylindole (DAPI), cells were cultured with 1 ml of DMEM (5.5 mM glucose) supplemented with 10% and sections were preserved using fluorescence mounting medium (Electron FBS. Post plating for 21 h, cells were washed twice with PBS and were subjected Microscopy Science). Images were acquired on a Nikon 90i Upright. ImageJ was 3–4 h to serum starvation with FBS-free DMEM (5.5 mM glucose). After washing used to process the images. twice with PBS, cells were cultured in 0.4 ml of glucose production buffer consisting of glucose-free DMEM (pH 7.4) without phenol red supplemented with 20 mM sodium lactate, 2 mM sodium pyruvate, 2 mM L-glutamine and 15 mM HEPES58. Palmitate and acetate oxidation assays. Seahorse Bioscience XFe96 extracellular Cells were incubated at 37 °C for 4.5 h with or without Bt-cAMP or pCPT-cAMP. Flux Analyzers were used53 to detect palmitate oxidation in primary hepatocytes. Both medium and cells were collected. The glucose concentration was measured Palmitate oxidation was measured by oxygen consumption rate (OCR) with with the Autokit Glucose (WAKO) and was normalized by the total protein modification. Primary hepatocytes were seeded at a density of 6000 cells per well of content. a XFe96 cell culture microplate and incubated in William E supplemented with 10% FBS. Next day, cells were cultured with DMEM (5.5 mM glucose) supplemented with 10% FBS and 2 mM Glutamax for 16 h. Then, these cells were Mitochondrial isolation. Liver tissue samples were minced and kept in ice-cold incubated in DMEM (5.5 mM glucose) supplemented with 1% FBS, 1 mM Gluta- PBS containing proteinase inhibitor immediately after harvest. Tissues were then max, and 0.5 mM carnitine for 2 h and then equilibrated for 1 h in palmitate homogenized in resuspension buffer (RSB)/EDTA (10 mM Tris pH 6.7, 10 mM oxidation assay medium (111 mM NaCl, 4.7 mM KCl, 1.25 mM CaCl2, 2 mM NaCl, 0.1 mM EDTA pH 8.0) containing proteinase inhibitor. The homogenized fi fi MgSO4, 1.2 mM NaH2PO4) supplemented with 2.5 mM glucose, 0.5 mM carnitine, samples were ltered through 30-μm lter and sucrose concentration was then and 5 mM HEPES at 37 °C for 1 h. 15 min prior to the assay, additional 400 μM of adjusted to 250 mM by adding 2 M sucrose. The suspension was centrifuged at Etomoxir or vehicle was added. Palmitate-BSA or BSA was added to the microplate 540 G for 3 min at 4 °C and the supernatant was collected for further separation. just prior to starting the assay. Sequential injections of 2 μM oligomycin, 2 μM Crude mitochondrial for functional analysis were sedimented from the supernatant phenylhydrazone, and 1 μM rotenone/1 μM antimycin A1 were used to examine by centrifugation at 9650 × g for 10 min at 4 °C. To prepare pure mitochondria, the mitochondrial oxidative status. crude mitochondria were resuspended in ice-cold separation buffer, mixed with For acetate oxidation assay, sodium acetate (1 M, pH 7.4) was prepared anti-TOM22 MicroBeads and enriched on a MACS column (Miltenyi Biotec). followed by filtering. Sequential injection of 5 mM acetate, 2 μM oligomycin, 2 μM Magnetically purified mitochondria were incubated with 100 μg ml−1 of RNase for phenylhydrazone and 1 μM rotenone/1 μM antimycin A1 were used to examine 30 min on ice and 10×volume of T10E20/sucrose was used to wash the mito- mitochondrial oxidative status. Each readout was normalized to total cellular chondria. Isolated mitochondria were pelleted and kept in −80 °C until use. protein levels. Luciferase assay. Luciferase plasmids harboring the Ampka1 3′ UTR were gen- Pyruvate oxidation assay and glycolysis stress test. Primary hepatocytes were erated as follows. Ampka1 3′ UTR were amplified by using primer: 5′-CCCAG isolated and seeded at a density of 6000 cells per well of a XFe96 cell culture AATTCCATTTAAGTTACAGCCTG-3′ and 5′-GCATCTCGAGGTTCCTTTC microplate and incubated in William E supplemented with 10% FBS. Next day, ATGAGAAATCAAC-3′, and cloned into EcoRI and XhoI restriction enzyme sites cells were cultured with DMEM (5.5 mM glucose) supplemented with 10% FBS. of pEZX-MT06 (GeneCopoeia). The EcoRI and XhoI sites are shown in italics. Next day, prior to performing an assay, growth medium in the wells of XF cell PCR was performed using Phusion High-Fidelity DNA polymerase (New England plate was exchanged with 175 μl of XF base medium (pH 7.4) containing no BioLabs). Mutagenesis of the 3′ UTR was performed with the QuickChange

NATURE COMMUNICATIONS | (2018)9:3658 | DOI: 10.1038/s41467-018-05870-6 | www.nature.com/naturecommunications 13 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-05870-6

Lightining Site-Directed Mutagenesis Kit (Agilent Technologies) according to the 9. Yecies, J. L. et al. Akt stimulates hepatic SREBP1c and lipogenesis through manufacturer’s instructions. Primer sequences used to mutate the miR-148/miR- parallel mTORC1-dependent and independent pathways. Cell Metab. 14, 152 binding site in the Ampka1 are the following: 5′-CATGATAGCTTGCAT 21–32 (2011). AAAAGATGACGCTATAGTTTAACGTCTGATTTCCGGACAAAAATG-3′ and 10. Arner, P. & Kulyte, A. MicroRNA regulatory networks in human adipose 5′-CATTTTTGTCCGGAAATCAGACGTTAAACTATAGCGTCATCT tissue and obesity. Nat. Rev. Endocrinol. 11, 276–288 (2015). TTTATGCAAGCTATCATG-3′. The mutated residues are indicated in bold. 11. Ross, S. A. & Davis, C. D. The emerging role of microRNAs and nutrition in Mouse primary hepatocytes were plated at a density of 0.8 ×105 cells per well of modulating health and disease. Annu. Rev. Nutr. 34, 305–336 (2014). 24-well plate, and transfected with 0.5 μg of each control (pEZX-MT06), Ampka1 12. Trajkovski, M. et al. MicroRNAs 103 and 107 regulate insulin sensitivity. (Prkaa1) (MmiT024101-MT06) and Ampka1 mutant using Lipofectamine 3000 Nature 474, 649–653 (2011). (Life Technologies) according to the manufacturer’s instructions. The transfected 13. Kornfeld, J. W. et al. Obesity-induced overexpression of miR-802 impairs cells were incubated with DMEM (5.5 mM glucose) with 10% FBS media for 8 h. glucose metabolism through silencing of Hnf1b. Nature 494, 111–115 (2013). Luciferase activities were measured with Luc-Pair Duo-Luciferase Assay Kit 2.0 14. Wagschal, A. et al. Genome-wide identification of microRNAs regulating (GeneCopoeia) and GLOMAX 96 Microplate Luminometer (Promega). cholesterol and triglyceride homeostasis. Nat. Med. 21, 1290–1297 (2015). 15. Goedeke, L. et al. MicroRNA-148a regulates LDL receptor and ABCA1 Polysome profiling. Preparations of cellular extracts for polysome profiles, sucrose expression to control circulating lipoprotein levels. Nat. Med. 21, 1280–1289 gradient centrifugation, and profile recording59. Heparin was omitted from lysates (2015). used to analyze Ampka1, Cs, and β-actin mRNAs by qRT-PCR due to its inhibitory 16. Soronen, J. et al. Novel hepatic microRNAs upregulated in human effect on the PCR. Sucrose gradient fractions were collected by upward displace- nonalcoholic fatty liver disease. Physiol. Rep. 4, e12661, 1–14, (2016). ment, and 50 pg of synthetic luciferase mRNA (Promega) and 15 μg of GlycoBlue 17. Ipsaro, J. J. & Joshua-Tor, L. From guide to target: molecular insights into (Life Technologies) were added to each fraction to control for extraction and PCR eukaryotic RNA-interference machinery. Nat. Struct. Mol. Biol. 22, 20–28 efficiency and to improve RNA recovery, respectively. After extraction and pre- (2015). cipitation of RNA from sucrose fractions59, precipitated RNA was washed twice 18. Meister, G. Argonaute proteins: functional insights and emerging roles. Nat. with ice-cold 70% ethanol, dried, and resuspended in RNase-free H2O. Equal Rev. Genet. 14, 447–459 (2013). volumes of RNA from each fraction were subjected to cDNA synthesis and qRT- 19. Winter, J., Jung, S., Keller, S., Gregory, R. I. & Diederichs, S. Many roads to PCR analysis. Levels of Ampka1, Cs, and β-actin mRNAs in each fraction were maturity: microRNA biogenesis pathways and their regulation. Nat. Cell Biol. normalized to luciferase mRNA and plotted as the percentage of total mRNAs from 11, 228–234 (2009). all 12 fractions. 20. Bartel, D. P. MicroRNAs: target recognition and regulatory functions. Cell 136, 215–233 (2009). Statistical analysis. Experimental results were shown as the mean ± SEM. The 21. Cheloufi, S., Dos Santos, C. O., Chong, M. M. & Hannon, G. J. A dicer- mean values for biochemical data from each group were compared by Student’s t- independent miRNA biogenesis pathway that requires Ago catalysis. Nature test. Comparisons between multiple time points were analyzed using repeated- 465, 584–589 (2010). measures analysis of variance, two-way ANOVA. Analyses of miRNA sequencing 22. Yang, J. S. et al. Conserved vertebrate mir-451 provides a platform for Dicer- were done by DESeq260. independent, Ago2-mediated microRNA biogenesis. Proc. Natl Acad. Sci. USA 107, 15163–15168 (2010). Data availability. The RNA-Seq data generated in this study have been deposited 23. O’Carroll, D. et al. A Slicer-independent role for Argonaute 2 in in the NCBI Sequence Read Archive (SRA, http://www.ncbi.nlm.nih.gov/sra) under hematopoiesis and the microRNA pathway. Genes Dev. 21, 1999–2004 (2007). the BioProject ID PRJNA395686. The accession codes are SAMN07415658, 24. Liu, J. et al. Argonaute2 is the catalytic engine of mammalian RNAi. Science SAMN07415659, SAMN07415660, SAMN07415661, SAMN07415662, and 305, 1437–1441 (2004). SAMN07415663 for the NCD study, SAMN07415764, SAMN07415766, 25. Valdmanis, P. N. et al. Expression determinants of mammalian argonaute SAMN07415768, SAMN07416029, SAMN07415767, SAMN07415769, proteins in mediating gene silencing. Nucleic Acids Res. 40, 3704–3713 (2012). SAMN07415765, SAMN0741630, SAMN07416036, SAMN07416037, 26. Ha, M. & Kim, V. N. Regulation of microRNA biogenesis. Nat. Rev. Mol. Cell SAMN07416038, and SAMN07416039 for the S961 study, and SAMN07413900, Biol. 15, 509–524 (2014). SAMN07413901, SAMN07413906, SAMN07413907, SAMN07413949, and 27. Carrer, M. et al. Control of mitochondrial metabolism and systemic energy SAMN07413951 for the HFD study. All other data supporting the findings of this homeostasis by microRNAs 378 and 378*. Proc. Natl Acad. Sci. USA 109, study are available from the corresponding author on reasonable request. 15330–15335 (2012). 28. Lu, L. et al. MicroRNA-130a and -130b enhance activation of hepatic stellate cells by suppressing PPARgamma expression: a rat fibrosis model study. Received: 2 February 2018 Accepted: 26 July 2018 Biochem. Biophys. Res. Commun. 465, 387–393 (2015). 29. Huang, J. Y. et al. MicroRNA-130a can inhibit hepatitis B virus replication via targeting PGC1alpha and PPARgamma. RNA 21, 385–400 (2015). 30. Park, J. E. et al. Dicer recognizes the 5’ end of RNA for efficient and accurate processing. Nature 475, 201–205 (2011). 31. Tsutsumi, A., Kawamata, T., Izumi, N., Seitz, H. & Tomari, Y. Recognition of References the pre-miRNA structure by Drosophila Dicer-1. Nat. Struct. Mol. Biol. 18, 1. Smith, B. W. & Adams, L. A. Nonalcoholic fatty liver disease and diabetes 1153–1158 (2011). mellitus: pathogenesis and treatment. Nat. Rev. Endocrinol. 7, 456–465 (2011). 32. Liu, Y. P. et al. Mechanistic insights on the Dicer-independent AGO2- 2. Perry, R. J., Samuel, V. T., Petersen, K. F. & Shulman, G. I. The role of hepatic mediated processing of AgoshRNAs. RNA Biol. 12, 92–100 (2015). lipids in hepatic insulin resistance and type 2 diabetes. Nature 510, 84–91 33. Kozomara, A. & Griffiths-Jones, S. miRBase: annotating high confidence (2014). microRNAs using deep sequencing data. Nucleic Acids Res. 42, D68–D73 3. Durnin, J. V. G. A. Basal metabolic rate in man.In Joint FAO/WHO/UNU (2014). Expert Consultation on Energy and Protein Requirements(Rome, 5–17 34. Gerlach, D., Kriventseva, E. V., Rahman, N., Vejnar, C. E. & Zdobnov, E. M. October). http://www.fao.org/docrep/meeting/004/m2845e/m2845e00.htm miROrtho: computational survey of microRNA genes. Nucleic Acids Res. 37, (FAO/ WHO/UNU, 1981). D111–D117 (2009). 4. Buttgereit, F. & Brand, M. D. A hierarchy of ATP-consuming processes in 35. Welter, D. et al. The NHGRI GWAS Catalog, a curated resource of SNP-trait mammalian cells. Biochem. J. 312, 163–167 (1995). associations. Nucleic Acids Res. 42, D1001–D1006 (2014). 5. Brown, G. C. Control of respiration and ATP synthesis in mammalian 36. Harley, J. B. et al. Transcription factors operate across disease loci, with mitochondria and cells Biochem. J. 284,1–13 (1992). EBNA2 implicated in autoimmunity. Nat. Genet. 50, 699–707 (2018). 6. Khamzina, L., Veilleux, A., Bergeron, S. & Marette, A. Increased activation of 37. Del-Aguila, J. L. et al. Genome-wide association analyses suggest NELL1 the mammalian target of rapamycin pathway in liver and skeletal muscle of influences adverse metabolic response to HCTZ in African Americans. Pharm. obese rats: possible involvement in obesity-linked insulin resistance. J. 14, 35–40 (2014). Endocrinology 146, 1473–1481 (2005). 38. Schaffer, L. et al. A novel high-affinity peptide antagonist to the insulin 7. Fu, S. et al. Polysome profiling in liver identifies dynamic regulation of receptor. Biochem. Biophys. Res. Commun. 376, 380–383 (2008). endoplasmic reticulum translatome by obesity and fasting. PLoS Genet. 8, 39. Perry, R. J. et al. Acetate mediates a microbiome-brain-beta-cell axis to e1002902 (2012). promote metabolic syndrome. Nature 534, 213–217 (2016). 8. Oyadomari, S., Harding, H. P., Zhang, Y., Oyadomari, M. & Ron, D. 40. Lu, Y. C. et al. ELAVL1 modulates transcriptome-wide miRNA binding in Dephosphorylation of translation initiation factor 2alpha enhances glucose murine macrophages. Cell Rep. 9, 2330–2343 (2014). tolerance and attenuates hepatosteatosis in mice. Cell Metab. 7, 520–532 41. Foretz, M., Guigas, B., Bertrand, L., Pollak, M. & Viollet, B. Metformin: from (2008). mechanisms of action to therapies. Cell Metab. 20, 953–966 (2014).

14 NATURE COMMUNICATIONS | (2018)9:3658 | DOI: 10.1038/s41467-018-05870-6 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-05870-6 ARTICLE

42. Burkewitz, K., Zhang, Y. & Mair, W. B. AMPK at the nexus of energetics and 60. Anders, S. & Huber, W. Differential expression analysis for sequence count aging. Cell Metab. 20, 10–25 (2014). data. Genome Biol. 11, R106 (2010). 43. Morita, S. et al. One Argonaute family member, Eif2c2 (Ago2), is essential for development and appears not to be involved in DNA methylation. Genomics 89, 687–696 (2007). Acknowledgements 44. Wang, D. et al. Quantitative functions of Argonaute proteins in mammalian We thank Drs. James Wells, Harinder Singh, David A D’Alessio, and Vivian Hwa for development. Genes Dev. 26, 693–704 (2012). their scientific input and discussion. We also thank Drs. Rebekah Karns, Taishi Kimura, 45. Lykke-Andersen, K. et al. Maternal Argonaute 2 is essential for early mouse and Kahori Ikeda for their bioinformatics and technical supports and Dr. Eric Lai for development at the maternal-zygotic transition. Mol. Biol. Cell. 19, 4383–4392 generously providing us Ago2-deficient fibroblasts. This work was supported by National (2008). Institute of Health (NIH) (R01DK107530). T.N. was supported by the PRESTO from the 46. Su, H., Trombly, M. I., Chen, J. & Wang, X. Essential and overlapping Japan Science and Technology Agency. A part of this study was supported by grants from functions for mammalian Argonautes in microRNA silencing. Genes Dev. 23, NIH (R01DK099222) to S.D. and P30DK078392 for the Digestive Disease Research Core 304–317 (2009). Center in Cincinnati. 47. Fu, S. et al. Aberrant lipid metabolism disrupts calcium homeostasis causing liver endoplasmic reticulum stress in obesity. Nature 473, 528–531 Author contributions (2011). T.N. conceived the research. C.Z., T.H., M.T.W., S.D., N.R.Q., C.A.M., H.S. and T.N. 48. Crescenzo, R. et al. A possible link between hepatic mitochondrial dysfunction designed the methodology and experiments. C.Z., J.S., K.M., E.S.B.S., E.B., V.J.B., K.C., and diet-induced insulin resistance. Eur. J. Nutr. 55,1–6 (2016). C.L.Y., C.C.C., X.C., N.R.Q., H.E.T. and T.N. conducted experiments. C.Z. and T.N. 49. Singh, R. et al. Autophagy regulates lipid metabolism. Nature 458, 1131–1135 wrote the manuscript with inputs from all authors. (2009). 50. Yang, L., Li, P., Fu, S., Calay, E. S. & Hotamisligil, G. S. Defective hepatic autophagy in obesity promotes ER stress and causes insulin resistance. Cell Additional information Metab. 11, 467–478 (2010). Supplementary Information accompanies this paper at https://doi.org/10.1038/s41467- 51. Madiraju, A. K. et al. Metformin suppresses gluconeogenesis by 018-05870-6. inhibiting mitochondrial glycerophosphate dehydrogenase. Nature 510, 542–546 (2014). Competing interests: The authors declare no competing interests. 52. Jonas, S. & Izaurralde, E. Towards a molecular understanding of microRNA- mediated gene silencing. Nat. Rev. Genet. 16, 421–433 (2015). Reprints and permission information is available online at http://npg.nature.com/ 53. Giles, D. A. et al. Thermoneutral housing exacerbates nonalcoholic fatty liver reprintsandpermissions/ disease in mice and allows for sex-independent disease modeling. Nat. Med. 23, 829–838 (2017). Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in 54. Stemmer, K. et al. Thermoneutral housing is a critical factor for immune published maps and institutional affiliations. function and diet-induced obesity in C57BL/6 nude mice. Int. J. Obes. 39, 791–797 (2015). 55. Arruda, A. P. et al. Chronic enrichment of hepatic endoplasmic reticulum- mitochondria contact leads to mitochondrial dysfunction in obesity. Nat. Open Access This article is licensed under a Creative Commons Med. 20, 1427–1435 (2014). Attribution 4.0 International License, which permits use, sharing, 56. Zhang, C., Seo, J. & Nakamura, T. Cellular approaches in investigating adaptation, distribution and reproduction in any medium or format, as long as you give Argonaute2-dependent RNA silencing. Methods Mol. Biol. 1680, 205–215 appropriate credit to the original author(s) and the source, provide a link to the Creative (2018). Commons license, and indicate if changes were made. The images or other third party 57. Rooney, J. P. et al. PCR based determination of mitochondrial DNA copy material in this article are included in the article’s Creative Commons license, unless number in multiple species. Methods Mol. Biol. 1241, 23–38 (2015). indicated otherwise in a credit line to the material. If material is not included in the 58. Sakai, M. et al. CITED2 links hormonal signaling to PGC-1alpha article’s Creative Commons license and your intended use is not permitted by statutory acetylation in the regulation of gluconeogenesis. Nat. Med. 18,612–617 regulation or exceeds the permitted use, you will need to obtain permission directly from (2012). the copyright holder. To view a copy of this license, visit http://creativecommons.org/ 59. Teng, T., Mercer, C. A., Hexley, P., Thomas, G. & Fumagalli, S. Loss of tumor licenses/by/4.0/. suppressor RPL5/RPL11 does not induce cell cycle arrest but impedes proliferation due to reduced ribosome content and translation capacity. Mol. Cell Biol. 33, 4660–4671 (2013). © The Author(s) 2018

NATURE COMMUNICATIONS | (2018)9:3658 | DOI: 10.1038/s41467-018-05870-6 | www.nature.com/naturecommunications 15